How to use xhuman

Start here

Email

How to

1. Connect a mailbox first if you haven't (ask me to “connect Gmail”).
2. Tell me who to write to and what to say — I draft it and show a Send card you approve.
3. Once I've shown you drafts in full — every word and who each goes to — one yes to them is enough: they all go exactly as shown, however many, with no more cards. A draft I changed after showing it still asks.
4. Ask me to find or summarize mail (“what did Sarah say about the contract?”) and I search your inbox.
5. Read receipts are on by default: mail you send from /mail, a Send card or a draft you okayed in chat carries an invisible pixel, and the sent message then shows when it was opened — and by whom when it went to one person (with several recipients, one pixel can't tell them apart). Your own views inside xhuman are ignored. Turn it off in Settings › Email & calendar, under Writing. A fetch by a security gateway or Apple Mail Privacy Protection reads as “delivered”, never as opened — those fetch on delivery, so treating them as a read would be wrong.
6. The “Sent via xhuman” footer is on by default, like Superhuman's: one line under your signature, linking to your referral link, so a signup from it counts as your referral. It goes under what you write in /mail (new messages, replies, forwards), compose cards, scheduled messages, Needs you and Inbox Zero — never a draft I showed you in chat that went on your yes there (it goes exactly as shown), and never campaign messages, not even a reply inside a campaign conversation. The composer shows the line under the body first. Remove it with the “Sent via xhuman” switch in Settings › Email & calendar; messages already delivered keep theirs.

Try asking

• Email the four people on this thread and loop in my cofounder
• Draft a reply to the latest message from Acme
• Find every email I have with jane@acme.com
• Did Ana open the email I sent her?
• How do I remove the “Sent via xhuman” footer?

Reading your mail

How to

1. Ask “what's going on with Greg?” or “did I already reply to Dana?” — I read your conversations with that person in every mailbox, newest first, and tell you who wrote last and when. If two people share the name, I ask which one.
2. Ask “what needs my reply?” — I list the replies you owe (campaign replies included) and your own asks that went unanswered past their date, each with the ask quoted and dated.
3. Ask me to open a conversation and I read every message in it. Name a mailbox (“in my Outlook”) to keep to that one; a rule like “only use my work inbox” keeps me to it.
4. A reply I'm holding for you is marked as a draft, never as sent, and mail that went out through xhuman says so.
5. If a mailbox couldn't be read — it needs reconnecting, or the provider failed — I name it, and never call your mail clear.

Try asking

• What's going on with Greg?
• Did I already reply to Dana?
• What needs my reply today?
• Read me the whole thread with Acme about pricing

Sign-in codes

How to

1. They're hidden wherever I read your mail: the email and /mail tools in chat, your tasks and suggestions, programs I run for you, your daily briefing and meeting briefs, the replies I draft (inbox drafts, replies your rules send, automation emails, forwards I start), guest preps, the rewrite in your composer, and your mail through the MCP connector or the voice interview. Where one was, I see a note saying a code or link was hidden, and I'll tell you to open the email.
2. I can't read one out, forward it, paste it into a reply or type it into a website — not even when you ask, and not in an automation: an automation reply to an email that carried one waits as a draft for you instead of sending. Open the email in /mail or your mail app to use the code yourself. The one exception isn't me: during a browser task, the browser itself can type a code a site emailed you into THAT site's page — I never see it.
3. A number or link I couldn't check in time is hidden the same way, marked “could not be checked” — it may just be a reference number or a year. Open the email to see it.
4. Two places read the words as written, on purpose: your CAMPAIGN conversations (a prospect replying to your outreach — the campaign reply drafter and its checks), and the ghost-text suggestions in your own composer, which only ever land in your draft. Your own sent messages are left as you wrote them.

Try asking

• Can you read my 2FA code?
• What's the verification code Google just sent me?
• Why can't you see the password reset link in my email?
• Forward my login code to my assistant

Safety screen

How to

1. In the chat, when something I read had such text in it, the tool card shows “Safety screen · removed 1 instruction aimed at the assistant” — open it to see exactly what was taken out. I see a marker where it was and never follow it, and I don't bring it up unless you ask about that content.
2. In an automation's run, the whole email or item that carried it is held back from that run instead, and the run doesn't describe it to you; if leaving it out leaves part of what you asked unanswered, it says that part is missing. Background runs that only propose (when I wake up after a task finishes, or the opportunity engine's sweep) get the text taken out the same way as in the chat, rather than the whole item held back.
3. Anything I couldn't check in time reaches me marked unverified, and I read it as data only — never as instructions.
4. It only takes out text aimed at an assistant: a colleague asking you to do something, a newsletter, or an article about AI reads normally.
5. The same screen covers a forward, a photo or PDF you send me, mail to my own address, and the other side of a thread you CC'd me on.
6. An automation that answers mail on its own holds the answer as a draft for you when the email tried to instruct me, or the answer says something your rule never gave it; its calendar invites go only to the sender and the people you named.
7. What it doesn't cover yet: my inbox agent (the one that reads each new email as it arrives) and the other always-on assistants that work through their own connection to your account, and the voice interview — those are read under their own rules, not screened line by line, and when a background run reads outside content, anything it then wants to change waits for your approval where that protection is on.

Try asking

• Did an email try to trick you?
• What does the Safety screen chip on a tool card mean?
• Can a web page give you instructions?
• Why does it say an instruction aimed at the assistant was removed?

Second look

Not available yet. This is not switched on for accounts today, so its steps are held back. This note clears itself when it is.

Fact check

How to

1. On an email card, the fact check says something only when a claim goes against your own records or calendar — “One thing in this email doesn't match what I found:” — and quotes it with what your records say. Edit it, or send it as it is — your call. A line nothing backs is fixed before you see the card, in my own words or by leaving it out, so the card you send is the fixed email; words you told me to say are never changed. A card with no line is not a guarantee.
2. There are no approved company answers yet, so a claim about how your company works — data handling, security, customers, pricing — can't be backed unless you said it yourself, in this chat or in your own email on the thread. The safer reply is “let me get you the exact answer.”
3. A line about what xhuman itself does — in an intro to xhuman, say — is checked against xhuman's own docs, the guides at /docs: a true one stays in your email, and one the docs don't say is fixed before you see the card, like any other line. A feature that's switched off right now doesn't count as something xhuman does.
4. If the email asks to book a call your calendar or recordings say already happened, or offers a time your calendar says you're busy, the card says your records say otherwise and quotes them.
5. For an automation that replies on its own: a reply with a claim your records or calendar prove wrong is held as a draft in that run, with that claim named, instead of being sent. A line nothing backs is trimmed or cut before the reply goes, never reworded with anything new; if it can't be fixed cleanly, or the check couldn't run, the reply goes as written — the run's record says what was and wasn't checked. Approving a held reply sends from the mailbox the email arrived in, Gmail or Outlook.
6. A reply alert with a one-tap Send shows the whole draft, so you read every word before you tap, and it keeps the one-tap unless a claim in the draft is proven wrong by your records or calendar. A draft too long to show whole in the alert gets a one-tap only when the check cleared every claim in it; otherwise you open it in /mail and send it from there.
7. A draft I write for you right in our chat, to paste or look over, gets the same fact check when the check can run: a line nothing backs is fixed in my own words before you see it, or left out, and a draft with nothing worth sending left in it is left out whole rather than shown half-cut. A draft held back because you turned on draft-only isn't rewritten, and nothing is added under any draft — so no note doesn't mean it was checked.
8. By text, Telegram, iMessage or WhatsApp, the confirm mentions the fact check only for a claim proven wrong, in one sentence before the email — “One thing in this email doesn't match what I found: … goes against your records.” What you told me to say — in your own words or with its plain consequence — and who you are (your profile) back a claim, so they are never flagged.

Try asking

• Why does the email card say a claim doesn't match what you found?
• Do you fact check the claims in my emails?
• Why was my automation's reply held — a claim doesn't match my records?

Automations on Just run it

How to

1. Turn it on in Settings › How it works for you under Automations (for all of them, or for one automation), or ask me — your ask turns it on, with no card — in the app, or typed in any chat, a plain text included.
2. What still waited says why on its card under “Waiting for you”.
3. A resumed campaign sends what it has due right away, with no delay of its own. Other campaign changes, like adding people, go ahead on their own after a 10-minute window you can stop — you're told on your phone and it shows on your list first.
4. Turn it back down any time on the same page or by telling me “make my automations ask me first” — it applies at once. Stop everything holds them too while it's on.
5. A change that goes ahead after the 10-minute window says so first: on your list “Goes ahead on its own in ~10 min” with Stop and Do it now; on Telegram Keep it for me and Do it now; by text just reply “hold” to keep it or “go” to do it now — with its number when more than one thing is waiting, as the text says (“hold” alone keeps every one). Stop means it waits for you on your list — nothing is dismissed.
6. Right before it goes ahead it checks again: if you pressed Stop everything or turned automations down since, it waits for you instead, and if it can't go ahead you're told. If going ahead on its own is ever switched off, those changes wait for you on your list instead.
7. If you've muted what it starts on its own (Settings › How it works for you), it can't tell you first, so those changes wait under Needs you on Home and Goals until you approve each one. Unmuting doesn't send them.

Try asking

• Can my automations just do it without asking me?
• Let my automations act on their own
• Why did my automation resume my campaign by itself?
• Stop that change before it happens

Automation browsing

How to

1. It's on unless you turn Websites Off in Settings › How it works for you (or ask me); Off applies at once.
2. Acting on its own needs your own words for that work, not a setting: “every Monday restock my coffee beans on bluebottle.com” lets that automation's browser task check out, and the payment asks you once with the total. A post, send or sign-up it meets along the way goes; a purchase your words didn't ask for asks you once, with the price.
3. A task an automation starts that could act goes ahead even when your words didn't ask for it, on a site set to Just run it: every button that moves no money just goes, and a payment goes under your purchase limits (over one, it asks you once, with the price). Anywhere else, each button that acts asks you first.
4. One site at a time, you can let a task YOU ask for start without its card: say “let it browse amazon.com without asking” or set that site to On its own in Settings › How it works for you › Websites — it applies at once. A task you ask for already starts on your words with no card, even one that will buy, post, send or sign up, goes beyond that site, hands the browser to you or runs signed in to your account there: the one thing I ask is the price, once, at the step that pays, unless you already said yes to that price in our conversation. Turn it off for that site any time.
5. With Browsing turned Off, an automation that wants to browse tells you it's off and that you can turn it on in Settings › How it works for you › Websites, or ask in a conversation instead, and nothing waits for you.

Try asking

• Can my automations browse websites for me?
• Why is a browser task waiting for me?
• Let my automation check a website

Scheduled jobs that do what you asked

How to

1. Just ask: “every morning check my LinkedIn and text me what's new”. It looks with no card, and you hear from it once, with what's new.
2. Sign in to the site first (Settings › Vault › Add a login) if it needs you signed in — it then looks as you, only to read.
3. It never presses Connect, Follow, Like or Place order while only looking: it stops there and tells you. It never pauses to ask you or hands you the browser while you're not there.
4. Turn Websites Off in Settings › How it works for you and nothing browses on its own.
5. A check posts, sends or signs up on its own as it goes, with nothing to turn on, and it runs signed in to your account on a site you connected.
6. A check pays for something only if your words for it asked for a purchase (“get me one bag when the beans are back”), and only under your cap; otherwise a purchase asks you once with the price.
7. A check with no words of yours on it acts by itself on a site set to Just run it: every button that moves no money just goes, and a purchase goes under your cap.
8. A check buys only up to your cap — change it in Settings › How it works for you › Money (default $100 a purchase, $250 a day). Over that, it asks you once.
9. To stop a check acting, cancel the check, or turn Websites Off for that site in Settings › How it works for you.

Try asking

• Check my LinkedIn every morning and text me what's new
• Can a check run on its own without asking me?
• How much can it buy without asking?
• Can a check buy or sign up for me without asking?

Your computer

How to

1. Ask me in the chat for anything that needs a computer (“install ffmpeg and keep it”, “run my scraper in the background and tell me when it's done”): I run it there, and the installs, files and sign-ins stay for later chats.
2. Long work keeps going in the background past about 2 minutes, and I come back to the same chat when it finishes. I extend the 2 hour default limit while the work is progressing.
3. A dev server or web app I start can be served on a public port while the computer is awake (port); a file I make can be shared to the chat (share).
4. Send me a file in any chat (web, Telegram, iMessage, text, WhatsApp, or email to my address): it lands on my computer in from-chat/, in a folder for that day, the next time the computer runs.
5. A browser task's downloads land on my computer in downloads/, in a folder for that day, and a browser task can upload a file from my computer to a form (“upload my resume.pdf to the application on jobs.example.com”). An upload sends your file to that site, so I ask once first unless you already named the file and the site.
6. Watch and manage it at /settings/computer (Computer in Settings): its screen live, with Stop and Take over; Earlier, each session titled in my words and replayable on that screen, with See the commands; Files of any size; and Signed in on this computer, with Remove and Remove all. Sign-ins stay until you remove them.
7. When you ask for something, I run the commands it takes, including one that sends something in your name, spends money, makes something public or deletes for good, without asking you again.
8. With outside content in the chat (an email, a web page), instructions aimed at me inside it are removed before they reach me. Credentials and one-time codes in output are masked before I read them.
9. Ask what a program connected to (“what did that script talk to?”): when the network log is on, it lists the websites and servers each command reached, host names only, kept 7 days, under See the commands in each session under Earlier at /settings/computer.
10. A desktop task (documents, spreadsheets, slides, other GUI tools) has no internet. On a current computer your commands stay online while it runs, it reads your workspace files without changing them, and what it makes comes back as the files it names, in a folder of the task's own under from-desktop/. On an older computer the whole computer goes offline for it and it works on your files directly.
11. It is paused when idle, so nothing runs while you aren't asking. Each plan has a daily computer-time allowance: Plus 150 min, Pro 360, Max 960, Ultra 1440.

Try asking

• What's on your computer?
• Install ffmpeg on your computer and keep it
• Run my scraper in the background and tell me when it's done
• Keep a dev server running while you test
• How do I see what's running on your computer?
• Write a file in your workspace and pip install requests
• Save a file I can use later
• Where did the file I sent you go?
• Upload my resume from your computer to that job form

Documents

How to

1. Ask for it in the chat: “redline this contract: change the fee to $12 as tracked changes”, “build a budget spreadsheet and check the formulas”, “make a 6-slide deck on X”, “fill in this PDF form”. Attach the file, or tell me what to start from.
2. The result arrives as a file in the chat; all of them stay in the documents folder at /settings/computer to download again.
3. A Word edit is a tracked change: the original text is struck through and mine inserted under my name, so you can accept or reject each one in Word.
4. A spreadsheet is recalculated before I send it, and I tell you about any #REF! or #DIV/0! cells I couldn't fix.

Try asking

• Make a Word doc with tracked changes
• Build an Excel sheet with formulas
• Make a PowerPoint deck
• Fill in this PDF form

Claude & ChatGPT plans

How to

1. Ask in the chat: “use my ChatGPT plan”, “have Claude make a 6-slide deck on our Q3 numbers and email it to me”, “use Codex to fix the failing test in acme/api”.
2. The first time, sign in with your own account. Claude: open the claude.com link you get (your Claude subscription is the default), sign in, and paste the code it shows back into the card. ChatGPT: open the auth.openai.com link and enter the code shown with it — nothing to paste back. You get the card here and on your phone; the job you asked for starts as soon as you're in.
3. Ask “is my Claude connected?” and I check: whether it's signed in, and which of your claude.ai connectors or ChatGPT apps it can use.
4. It runs in the background; the result comes back in the same chat (with the file), and by email when you asked for that.
5. Remove the sign-in any time under Signed in on this computer at /settings/computer.

Try asking

• Use my ChatGPT plan for a job
• Connect my Claude subscription
• Have Claude make a presentation and email it to me
• Use my ChatGPT plan to research a topic
• Use Claude Code to fix a bug in my repo
• Continue the Claude Code session

Sites

How to

1. Ask me in the chat: “make a 90s movie trivia game I can share with friends”. I make the page and reply with what's in it and the link.
2. Want something to use, not just read? “Make me a page to track my workouts this month from my phone” — the entries stay.
3. Just tell me what goes on it: “add today's 5k run to my workout tracker” — I put it in for you, and change or remove entries the same way. No form to fill.
4. In this web chat your page shows live under my reply: add an entry right there, press Expand for the full height, or Open in a new tab.
5. Want it different? “Change question 3” or “add the prices” — the new version goes up at the same link, even from another chat.
6. Your pages are in Settings › Files: open one, copy its link, or take it down.
7. Anyone with the link can open a page. It never shows your private details unless you ask me to put them on it.
8. A folder on my computer can be published too: I send a card first, and your yes publishes a copy of it.

Try asking

• Make a trivia game I can share
• Make me a page I can send to friends
• Take down my page

Asking you

How to

1. When Claude Code or another program on your computer needs a login code from you, it arrives as a card on your phone. Answer a card by tapping a button, replying with the number of the choice, or typing the answer itself.
2. On the web, answer on the card in the chat. A secret opens a secure page instead, so it never goes through a message.
3. A sealed answer goes straight to the program that asked; I never see it, and it isn't saved in the conversation.
4. Anything working for you while you're away — me, or a program on your computer — can ask you something and wait; ask me to check in with you the same way. Your answer picks the work back up.

Try asking

• How do I give Claude Code the login code from my phone?
• Can my computer ask me something while I'm away?
• Where do I paste the sign-in code?

Your own Mac or PC

How to

1. To connect one, ask me for a link in the chat, or open Settings › Computer and press Connect a computer under Your computers. Open the link on the computer you want to connect.
2. On that computer, paste the line into Terminal (PowerShell on Windows) and press Enter. The helper says whose xhuman the computer joins and waits for you to press Enter again before it connects. The link works once, for 10 minutes.
3. On a Mac, it may show Background Items Added, and ask to let node use your Desktop, Documents or Downloads folder: click Allow.
4. Then ask me for anything on it by name, like "what's on my MacBook's desktop?" or "run the tests on my work PC". I can list your computers, online or asleep, and when each was last seen.
5. Your own Claude Code and Codex sessions on it: ask "what are my Claude Code sessions on my Mac?", "what did Codex say in the importer thread?" or "tell my Claude session in the api repo to also fix the signup test". I find the session by the name you see on it (archived ones too), pass it your message, and return its whole reply here, however long; if it is still working, I check on it again with a timer. A question you ask a session to answer goes to it; I don't answer in its place. It runs with that session's own settings; when it reaches a step those settings ask about, I tell you what it wants to do and ask; it does that one step only on your yes, and a no leaves it undone. It holds the step 30 minutes. Ask for a past session and I open it where it lives (on your computer, or in my browser for a web one I read in), with where it stopped and an offer to keep going. With no computer added yet, I send the link to add one. If the session is in use or the computer is asleep, I tell you and can check back; I send it when you say go. What I can't see as in use: a session opened without its name and sitting idle, a plain claude whose first answer runs past two minutes, or a computer whose process list can't be read (I say so). An open window shows my message after you reopen it; for a session open in its app, I type into that window instead and return its reply.
6. If the computer is asleep or offline, nothing runs anywhere else: I tell you. A command you asked for runs as soon as it's online again; a message to one of your sessions isn't sent until you say go.
7. On a Mac with the xhuman app and its two switches on (Accessibility and Screen Recording), I can also work in its apps, like adding a note in Notes, behind your windows. A task you asked for does every step it needs there, sending, deleting or buying included when that's what you asked, with no second question; before it pays I tell you the item and the total once (unless your ask already named them), and it pays only with a card or payment method already saved in that app on the computer you connected, never typing a card number. One an automation starts does every step too but never pays. If an app won't work behind your windows, I bring it to the front for that step, with no question.
8. When a site blocks my cloud browser, the same task moves to your connected computer by itself, if one is online, and I tell you it ran there. If none is connected, I offer you the one-time link to connect one, and the task waits for it. Tell me not to use your computer for a task and it stays in the cloud.
9. On a Mac with the xhuman app and its two switches on, I can take a screenshot of a window, like your coding session, or of your whole screen, and the picture comes back here, in the same chat.
10. A photo you give me here can go into an app on that Mac, like your coding session's message box: I paste it where you would, then type your words as asked. Your clipboard is put back right after, and my copy on the Mac is removed when the task ends. I can also save it into a folder there, like Downloads.
11. A command that sends, pays, posts or deletes for good still asks you first unless you asked for it, exactly as on my own computer. Disconnect, from Settings or by asking me, cuts the computer off at once; connecting it again takes a new link.

Try asking

• How do I connect my laptop to xhuman?
• Can you use my laptop?
• Connect my computer
• What's on my Mac's desktop?
• Disconnect my laptop
• Add a note in Notes on my Mac
• Message my Claude Code session on my Mac
• Send me a screenshot of the session open on my Mac
• Paste this photo into my coding session on my Mac

Skills

How to

1. If someone asks what skills I have or what I learned, point them to Settings › Skills: every skill is there, with what I follow for it, when I last used it, and Undo on my latest change.
2. If someone wants me to stop learning, the switch at the top of Settings › Skills turns it off; with it off I don't learn and I don't use the skills I have. Turning one skill off is the Turn off button on its row.
3. Can a skill you wrote send emails without asking me? A skill is only a how-to: it adds no permission. Sending, paying, posting and deleting go exactly the way they always do, with the same asks.

Try asking

• Can you learn new skills?
• Teach yourself how to do my weekly report
• What skills have you learned for me?
• Turn off my invoices skill
• Stop learning skills

Outside content safety

How to

1. Held actions wait under “Waiting for you” (the proposals rail on /orb) as “Queued action” cards that start “The agent wants to …” and lead with what approving does. Run it does exactly that one thing; dismissing it drops it. The same action asked for again while its card is open lands on that card, not a new one.
2. The card says why it waited — for example “Held because the automation run read an email someone sent you.” — and lists every detail it would run with, whole, so an instruction that came from the email rather than from you is easy to spot.
3. Only you can run a held action, with Run it on its card under “Waiting for you” on /orb — the one place that shows every detail it will run with. A message about it (Telegram, a text, the daily email) only points you there; it has no button that runs it. I can't — not even if you tell me “yes, do it” in the chat (I'll point you to the card), and a background run can't approve or dismiss it — asking again only lands on the same card. I can list what's waiting, tell you where it is, and drop one if you say no.
4. When running it means first writing something — an email, a LinkedIn message, a change a program works out — Run it writes that and puts it back on the card for you to read. Nothing is sent or saved until you approve what it wrote.
5. So an email can't trick an automation into emailing someone you didn't name, paying, publishing or deleting by itself: what goes on its own is only what your own typed words asked for, checked against them, and the most a planted instruction gets is a card in front of you, saying what it would do. When outside content is caught carrying an instruction aimed at me, nothing from that run goes without your yes. A clear-cut planted instruction is stopped outright; you get the one notice, and if it kept me from a step you asked for, I tell you that step wasn't done.
6. When you're in the chat yourself, none of this applies: things run as you ask — the reminder, the watch, the automation you asked for is set up right away, with no extra card, whatever the conversation read before, even when it held an instruction aimed at me (I get that text stripped out).
7. An automation I carry out from its instructions (a news watch included), a standing rule, or a saved enrichment or search criterion saved after reading outside content keeps the mark: when I later work from it in the background, I check with you before it reaches other people, spends money, publishes or deletes.
8. How long a read counts: while it is still in what the run can see. A run an hour or more after the last one sees earlier emails and pages only as summaries, and those stop counting; a held action stays in view, so the next runs keep asking first until it scrolls out of a long thread.
9. What this doesn't cover yet: my inbox agent (the one that reads each new email as it arrives) works under its own, separate rules — it drafts replies for you to review and notes facts, except where a rule you taught it lets it answer on its own: those answers go out after a few minutes in Sending soon, written from the email they answer; you can hold them there. And the words I write in a reply don't make a later run ask first on their own: if I summarized an email in an earlier run, that summary alone doesn't hold anything — but the quiet check remembers it, reads what a later run does next against what you asked, and stops it outright if it clearly came from outside. Nor is your notes wiki: an email or a web page you saved into it (+save, or a link you added) reads back as your own notes, so a run that reads it there doesn't ask first because of it.

Try asking

• What happens if my automation reads a phishing email?
• Can an email trick my automation into sending something?
• Why is my automation asking me to approve an action?
• Why is an automation waiting for me?

Relationship inbox

How to

1. Read-only on a connected account is enforced by us, not the provider: every write on it is refused on our side, so it never reaches them; the access you granted Google, Microsoft or the app at connect is unchanged. For a mailbox, ask — or say so when you connect it: I keep reading it and never send, draft or change anything in it.
2. Open /mail (both Gmail and Outlook mailboxes are indexed), or ask me “what needs a reply?” / “where do things stand with Sarah?” / “anything from Acme this week?”.
3. Tell me what you're working on (“I'm raising a seed round by December”) and I record it as a goal the Opportunities tab judges new mail against — Goals & Filters on the Opportunities tab lists them, with confirm and retire.
4. Say “add a filter for podcast invitations” or “catch emails from journalists asking for quotes” — I create the filter in your words, can try it on your 50 newest threads first, and from then on it runs on new mail as it arrives. Ask “what did my investor filter catch this week?” for its matches, and “notify me when it catches something” to get a message on every channel you have linked.
5. Ask me “which campaign replies still need my answer?” or “did I already reply to Nathan?” — the Campaigns tab lists every reply to your outreach with its state, and I can open the conversation and draft the answer.
6. Inside an email in /mail, the ∧ ∨ arrows at the top of it (or k / j) open the previous or next conversation in the list you opened it from — a campaign's To reply tab included — so you can answer one reply after another without going back to the list. E (Close, on a campaign reply) files the one you are on and moves on to the one below it.
7. CLEAR OUT A KIND OF MAIL: “clear out my promo mail” or “archive every newsletter older than a month, except my bank and airlines” — a dry run counts it by sender and by mailbox and lists what stays; on your yes I archive every match across every page of /mail and every connected mailbox, Gmail and Outlook alike. It archives, never deletes, and leaves starred conversations, threads you wrote in and replies to your outreach alone; “restore them” returns them all to the inbox. It covers what /mail holds (the last 90 days). “And from now on” saves it as a standing rule that archives new matching mail as it arrives (a rule for new mail takes no age); removing the rule ends it, and what it archived stays archived unless you have me restore it.
8. Ask me to mark a conversation done, snooze it, pin it, delete it, or say a reply is expected by Friday. Done and a reminder archive the conversation in your mailbox too, Gmail and Outlook alike (it leaves your Inbox there), and undo puts it back. Delete (# in /mail) moves it to Gmail's Trash or Outlook's Deleted Items, recoverable there; it never deletes for good. Read state and stars follow in Gmail and Outlook. Nothing is ever sent.
9. AUTO REMINDERS: when an email you send from /mail (a reply, a new message, or one you approve from chat) asks for an answer, I set a reminder on it myself — the conversation moves to ☰ → Reminders and comes back at 9:00 AM on the business day I picked from what you wrote (sooner for “by tomorrow”, later for “no rush”) if nobody replies; a reply brings it back at once. Its clock reads “Auto reminder”; h moves or removes it like any reminder, and Settings › Email & calendar › Auto reminders turns them off. A thank-you or an FYI gets none, and sensitive threads are never read for one.
10. Ask me to draft a reply (“say yes and propose Thursday”); I show you the draft, and sending waits for your approval — one yes sends every draft it covers, and it cannot be unsent.
11. YOUR FACTS IN DRAFTS YOU REVIEW: the /mail AI draft and an Inbox Zero card's first draft carry what you told me about yourself — pinned facts, for-now ones like out of office until a date, preferences — as background; the drafter is told never to raise one the conversation didn't, sensitive ones are dropped, and cold outreach or replies without your review (a campaign on Just run it, a rule that answers on its own) use none.
12. With /mail open, the side chat sees what you see: the conversation on screen, the person, the campaign a reply answers, and the reply you are writing. Say “fix the name in this email” or “make my reply shorter” and I rewrite the draft in your composer — the words change in place, nothing is sent, and ⌘Z in the body brings yours back.
13. Say “make a VC offer inbox” or “add an AI inbox for podcast invitations” and I create a label in your words and pin it to the top row as its own tab — every conversation it tags, beside Needs you and Opportunities; “unpin it” or “take it off the top row” turns it back into a plain label in the folders panel. Same label, only where you reach it from.
14. Say “label this Prospect” or “tag them as an investor” on a conversation and I put one of your own labels on it by hand — which also records it on that person's record in your network, so you can find everyone you have labelled that way and build a list from them. A label the classifier applies on its own stays on the conversation; only one you apply becomes a fact about the person.
15. To make a new label yourself, press l in /mail on a conversation (or on a selection) and type: the list filters as you type, Enter puts the highlighted label on (or takes it off when it is already there), and a name no label has yet shows “Create label …” — press Enter and the label is created and applied in one keystroke. A label made that way is one you apply by hand: it never runs on new mail until you describe it in the labels manager. z undoes it, and undoing a label you just created deletes the label too.
16. Ask me about any email or thread — “what's the thread with Dana about the demo?”, “find the email from Stripe about the refund” — and I find it in your mail and put that conversation under my answer with an Open thread button that takes you straight to it in /mail (in place, when /mail is already open).
17. SEARCH in /mail works like Superhuman's: press / (or the magnifier at the top) and the box opens on the person you're on. As you type, the people you mean appear — Tab puts their address in the box — and results update live, every matched word highlighted, with the passage around the match. Operators: from: / to: (a name or address), subject:, "exact words" in quotes, is:unread / is:starred / is:snoozed / is:done, in:inbox / in:done / in:sent (and -in:inbox), before: / after: (2026/06/01), older_than:3d / newer_than:1m — the Tips panel beside the results lists them. The first suggestion, “Ask AI”, hands your question to me (⌘↵ from the box). Esc steps back: from a result to the results, to the box, then out.
18. Ask “what did I promise Jane?” for the promise ledger in both directions.
19. Ask for the follow-ups from your meetings over a period — each promise you owe people or they owe you, from the calendar and the mail around each; a promise already kept says so.
20. Open ☰ → Drafts for the replies you started and never sent (clicking one reopens that draft), ☰ → Sent for everything you have written, and ☰ → Scheduled for the emails I am holding to send later — cancel, reschedule or send one now from there, or ask me “what's scheduled to go out?”.
21. The pencil at the top of /mail (or c) starts a new email to anyone, from whichever connected mailbox you pick.
22. Ask me to chase a quiet conversation (“if Ana is silent by Friday, follow up”): I set a wait on it. At the deadline I draft ONE follow-up for you to approve. Their reply cancels it and you hear once, with a quote; if I can't prepare it I say why.
23. SEND LATER, like Superhuman's: in any /mail composer — reply, reply all, forward or a new message — press ⌘L (⌘⇧L works too) or the button's ▾, then type when (“tomorrow 9am”, “in 2 hours”, “monday”, “oct 3 2pm”) or pick in an hour / tomorrow morning / tomorrow afternoon / Monday morning. The pick schedules it on the spot, and the notice after it has Undo. It waits in ☰ → Scheduled — cancel it, move it or release it early from there — and it shows in its conversation too, with Edit. If they write again before its time, it is held back as a draft for you to read first, and you are told; nothing retries on its own.
24. /mail opens on IMPORTANT (first on the top row, g i): every conversation in your inbox that your mail provider filed as mail from a person — Gmail's Primary tab, Outlook's Focused Inbox — newest first, whoever wrote last, the way Superhuman's Important works (calendar notices and senders you muted go to Other). Mail your provider never sorted (LinkedIn, older threads) is judged by xhuman instead: what needs you, replies to your outreach, threads you have written in, and people you have emailed back and forth with. A conversation stays on Important until you mark it done (e) — or archive it in Gmail, Outlook or Superhuman: within about 15 minutes it leaves here too, and it comes back if it returns to that inbox. Done here archives it in Gmail or Outlook too, so Superhuman agrees, and undo (z) puts it back. Needs you (g w) is the ranked list of what is waiting on your reply.
25. Everything that reaches your mailbox is in /mail — not only mail from people. The OTHER tab (last on the top row, g r) is the rest of your inbox, what Important did not claim: what Gmail files under Updates, Promotions, Social or Forums (Outlook's Other), newsletters, notifications, receipts and alerts, calendar notices, a stranger's pitch the judge passed on — the last 90 days of every category, which xhuman imports once per mailbox in the background. Needs you never lists mail your provider filed there, or a calendar notice (a receipt that asks a question is not a person waiting on you), unless you tell me it needs you. ☰ → Spam (g x) is what your mail provider flagged. ☰ → All mail is every conversation but spam, including what you archived. So if an email is in Gmail or Outlook, it is here somewhere; a row wears “newsletter”, “notification” or “spam” when that is why it is not on a tab. Showing Other or Spam calls no model — those rows are read from the index alone until you open one.
26. If a mailbox stops syncing — Gmail or Outlook stopped letting xhuman read new mail, or its access was revoked or expired — its row in Settings says “Not syncing” or “Reconnect”, and on a paid plan you are sent a reconnect notice (plus a Telegram or SMS ping if linked) with a Reconnect button: at most three per outage, three days apart, stopping once you reconnect (or choose Disconnect on that row, if you removed access on purpose). Reconnecting is signing in to that mailbox again; nothing already in /mail is lost.
27. Your mailbox's own state follows you here, one way, within about 15 minutes: a conversation you star in Gmail (or flag in Outlook) is starred here too (☰ → Starred), one you read there is read here, and a reminder you set in Superhuman — or Gmail's own snooze — moves the conversation to ☰ → Reminders, marked “in Superhuman”, until it comes back to your inbox. A send Superhuman marks as Waiting goes to Waiting on them, and I also read each plain email you send (one that asks nothing in so many words) to tell whether you are waiting on an answer. Your Gmail labels, Superhuman's topic labels (Marketing, Pitch…) and Outlook categories show up as labels here. The other way works too, within about a minute: file, delete, read, star or set a reminder on /mail and Gmail or Outlook follows. Labels stay in xhuman, and a star or label you take off here stays off.

Try asking

• What needs a reply in my inbox today?
• Where is my Important inbox?
• Why isn't this email in Important?
• Show me my drafts
• What have I sent this week?
• What emails are scheduled to go out?
• Catch emails from journalists asking for quotes
• Add a filter for podcast invitations and notify me
• What did my investor filter catch this week?
• Label this conversation Prospect
• How do I make a new label?
• Tag them as an investor
• Make a VC offer inbox
• Add an AI inbox for podcast invitations
• Tell my inbox I'm raising a seed round
• Which campaign replies still need my answer?
• Show me the Warm Reconnect replies I haven't answered
• Did I already reply to Nathan Bell?
• Open the thread with the person who wants to schedule
• What's the thread with Dana Whitfield?
• Find the email from Stripe about the refund and give me a link to it
• Where do things stand with Sarah Chen?
• Snooze the Acme thread until Monday
• Clear out my promo mail
• Archive every newsletter older than a month
• Archive marketing mail as it arrives from now on
• Why did my thread go to Reminders after I replied?
• Turn off auto reminders
• Draft a reply to Marco saying yes, and propose Thursday
• What did I promise people in last week's meetings?
• Where are my newsletters?
• Show me my spam
• An email I got in Gmail isn't showing in xhuman
• Can I see everything that arrived today?
• Why doesn't my Superhuman reminder show up?
• Do my Gmail labels show in xhuman?
• How do I search my mail?
• Search for unread emails from Dana

Edit with AI

How to

1. Select the words to change in a /mail reply or new message, an email card I draft in chat, or the campaign sequence editor, then press ⌘K or the Edit pill over the selection. In a campaign's step card, select the words and use the card's own “Edit selection” button.
2. Type what should change (“friendlier”, “add that Thursday works”) or tap a suggestion. A few words widen to their whole sentence, and the bar says so.
3. Improve writing makes it clearer and tighter and fixes grammar and spelling — same meaning, same facts, no more formal than you wrote it.
4. Rewrite in my voice rewrites it the way you write, learned from your own sent emails — how you write to this person when you have written to them before. What I have learned is in Settings › How it works for you, under “How it writes for you”; correct it there and every rewrite follows your version.
5. Undo puts your words back, Try again asks once more, Keep closes it. Nothing is sent — only the draft in front of you changes.
6. It works on up to about 2,000 characters at a time. A conversation locked from AI offers no Edit pill — its text never goes to a model. One marked sensitive asks for your OK first (give it once with AI draft); after that, edits read it like any other thread.

Try asking

• Rewrite this in my voice
• Improve my writing
• How do I make an email sound like me?
• Can the AI fix the grammar in my reply?
• How do I edit part of an email with AI?

Autocomplete

How to

1. It shows in /mail replies and new messages, and in the email cards I draft in chat. Not in campaign templates or sequence steps — those go to many people.
2. Pause for a moment and the suggestion appears in gray. Tab or → accepts it, Esc dismisses it, typing the same words keeps it, and on a phone you tap it. ⌘Z takes an accepted suggestion back.
3. It never makes up names, numbers, dates, links or promises — when it doesn't know, it suggests nothing. A conversation locked from AI gets no suggestions. One marked sensitive gets none until you have OK'd AI for it.
4. When your credits run out, suggestions simply stop, and your typing is never blocked.
5. Turn autocomplete off in Settings › Email & calendar › Writing › Suggest as I write. It is on for every subscription.

Try asking

• Can xhuman autocomplete my emails?
• How do I turn off autocomplete?
• Why is gray text showing up when I type?
• How do I accept the suggestion?

Auto labels

How to

1. Hover a chip on any conversation to see why it is there and who decided — the AI, a rule, you, or something xhuman knows. Keep it or remove it (“Not pitch”) in one click; from the keyboard, press l, then Enter to keep or ⇧Enter to remove. Every correction is remembered and teaches the label what you mean.
2. A reply to your own outreach is never labelled pitch — it wears its campaign instead.
3. Open ☰ → Auto labels to see every conversation a label is on, or the ⚙ beside the heading for the Labels manager: switch a label off, recolour it, or give it its own tab in the top row.
4. The Library (in the Labels manager) adds more — recruiting, investor, customer, press, legal, travel, security — one click each. Or build your own in a sentence and try it on your recent mail before you save it; mark a few matches ✓ or ✕ and they are saved with it as examples.
5. “Pause all labels” in the manager pauses them all. On for every subscription.
6. Labels show in Gmail and Outlook too, as xhuman/pitch, xhuman/intro… — so Superhuman and any other mail app show them. “Show in Gmail and Outlook” in the Labels manager turns that off; labels already written stay until you remove them there.

Try asking

• Why is this email labeled pitch?
• What auto labels does xhuman have?
• How do I turn off the marketing label?
• Can labels show which campaign a reply is from?
• Why was this email labeled newsletter?
• Do xhuman labels show up in Gmail?

Things to do from your mail

How to

1. Tap an item to see two or three sentences from the email and one button to where it lives — the invoice, the settings page, or the email itself.
2. Mark it Done when you've handled it, or Dismiss it if it isn't yours to do. Later emails about something you dismissed stay off your list.
3. Amounts, dates and names are checked against the email before they're shown. A button opens a page only when it is on the sender's own verified domain; otherwise it opens the email.
4. Sensitive mail — bank details, passwords, sign-in codes — is never read for this.
5. When it first turns on it looks back over your last 30 days of mail once, so Home isn't empty on day one; after that it reads new mail as it arrives. On for every subscription.

Try asking

• Where do the bills on my Home come from?
• Why is this invoice on my Home?
• How do I get rid of an item from my email?
• Does xhuman find bills in my Outlook too?

Inbox rules

How to

1. Ask me how your email is handled, or open Settings › How it works for you — the Email replies row lists what you taught me to do on my own and the people and domains you set apart, each with its undo.
2. To let me act on a kind of mail on my own, tell me in your words — “you can take care of scheduling for me”, “confirm things for me”, “decline cold pitches politely” — and it becomes a rule as I'll follow it, on from that moment: in the app, or typed in any chat, a plain text included. With something from outside already in view, I no longer show you its words once first. After you approve three of one kind unedited, I offer it once myself. A rule can name one mailbox.
3. To keep something waiting for you, tell me — “always draft for investors”, “leave recruiters to me” — and I save it as a rule I read before acting; it wins over any rule that lets me act. Pausing or removing a rule applies at once — a reply it had in Sending soon waits for you instead — and nothing already waiting is sent.
4. In /mail, opening a conversation that needs an answer puts its draft (“Drafted for you”) straight into your composer, ready for ⌘↩ to go out — yours to edit, and it waits for that key. Esc closes it and “Use it” brings it back; words you wrote there first are never replaced. The draft answers the whole conversation, and a burst of messages from one person gets one draft.
5. No draft? Usually it needs no answer — a cold pitch, a newsletter, an FYI, a thank-you or automated mail gets no draft and no ping. Otherwise a rule on the conversation, person, domain or label turned drafting off there (“Why?” says which), a campaign is handling that person, or it is sensitive (bank details, passwords, health), which is never drafted.
6. To stop drafting in one place: the Replies chip on a conversation has “Don't draft here”, a person's menu has “Don't draft replies”, and a label's ⚙ can turn drafting off for everything it catches. Or tell me — “stop drafting for dana@acme.com”, “stop drafting for everyone at acme.com”. Turning drafting back on there is yours alone: I only do it when you ask me yourself, never from an automation or a suggested question.
7. What a rule answers on its own: whoever wrote, someone you never wrote to included, but only what your own earlier words in the thread already cover, or a plain confirmation. Pricing, contracts, sensitive mail, group threads and anything your own words can't back up wait for you, drafted, with one sentence saying why.
8. Each answer goes out right away — or about 20 minutes later when one of your automations also answers email, so that one answers first — and one due in your quiet hours waits until they end. While it sits in Sending soon (in /mail's Scheduled) you can say “Don't send” or Edit there, or tell me to hold it; writing to them yourself in that conversation holds it too. Once it has gone it can't be unsent.
9. Handled in /mail lists what it did — Answered, Waits for you (with the reason), No reply needed (with an Undo) — and a message it sent carries “Sent by xhuman”; “what did you send for me today?” lists it here. On Telegram and SMS it tells you before each of its first 20 answers (by text, replying “hold” keeps it as a draft), what waits for you, and a morning recap of what it did the day before; with no messenger linked, the waiting list and the recap come by email — nothing is announced before it goes.
10. There's no daily limit: your inbox keeps sending the replies your rules cover, however many come in. We can switch answering on its own off for a while — answers then wait for you, and I tell you.
11. Following up: when people you wrote to yourself go quiet past the time you'd expect, I write ONE short follow-up in their language, and it waits for you, drafted (in /mail, in Needs you, and on your phone). There's nothing to turn on — following up is drafted for every mailbox. Never a second follow-up, never on mail I wrote for you, never on a group thread or someone a sequence is writing to; family and money, legal or contract threads wait for you.
12. Offering times and booking: asked to meet, I offer three free times from the calendar of the mailbox it came to (Google for Gmail, Outlook for Outlook), in their language, drafted for you; when they pick one, the confirmation is drafted too, and sending it yourself books the meeting — the invitation goes with it. If they share their own booking page instead, tell me you want it and I book a free slot on it for you — a time your calendar has free, with your own details — then tell you the time I booked; anything I write back to them just confirms it. With a rule that lets me book meetings for you, the offer goes out after Sending soon and a pick books itself — Sending soon says “Booking Thu 3pm with Ann”, and “Don't book” holds both; to stop it sending calendar invites on its own, pause or remove that rule. Moving or cancelling, their own booking link, a time I didn't offer or anything besides the meeting waits for you. If an invitation may have gone out but I can't tell, I never try again — I tell you to check your calendar.
13. In /mail, the first visit of the day shows a strip — “Since yesterday: answered 2 · followed up 1 · 1 waiting for you.” — and its link opens Handled (“See what I sent →” when something went out, “See what waits →” when drafts just wait).
14. A rule on one person, their company or a conversation also binds your campaigns — on Just run it they never write back to them on their own (nor book, share your booking link, move or cancel, or write the one follow-up after times went out); they draft it and wait for you. Their sequence still goes, as does their unsubscribe acknowledgement — pause or remove the person there for that. So “stop auto-replying to dana@acme.com” covers campaigns too.
15. Told to quit answering a group on its own — investors, say — I save a standing rule I read before every answer. Rules in your words bind mail outside campaigns only, so I say “Done for mail outside campaigns” — and if one of your campaigns writes to them on Just run it, I name it and offer to lower it to Approve replies: its own mode dial, which I can lower here.
16. It never changes your campaigns' modes — each keeps its own (its dial) — and automations follow their own rules.

Try asking

• How is my inbox handled?
• Let it take care of scheduling on its own
• You can take care of scheduling for me
• Always draft for investors
• Never answer recruiters
• Stop drafting replies to dana@acme.com
• Stop drafting for everyone at acme.com
• Can you auto reply to my emails?
• Do you auto-reply to emails for me?
• Why didn't you draft a reply to this email?
• Which emails will you draft replies for?
• Why is there no draft for this cold pitch?
• What did you send for me today?
• What did it do on its own today?
• Don't send that reply
• Why didn't it answer Carl?
• Stop auto-replying to investors
• Stop auto-replying to dana@acme.com
• Reply to cold pitches with a polite no
• Follow up with people who don't answer my email
• Can you follow up for me when people don't get back to me?
• Did you follow up with Ann?
• Turn on following up for my work mailbox
• Can it book meetings for me when someone picks a time?
• Why didn't it book the meeting with Ann?
• Keep calendar invites waiting for my approval

Work list

How to

1. COME BACK TO IT LATER: say "skip this one for now" or "come back to this after the rest" and I note it on this chat's work list. In Inbox Zero the deferred card returns after the others. The list only orders the work: it never marks mail answered, sends nothing and schedules nothing.
2. LIVE IN THE CHAT: on the web my plan sits under the newest message and ticks off as I go, no reload; when a job carries on after my reply, that same plan keeps updating there.
3. WHERE IT SHOWS: open Tasks — the chat's steps are nested under its task (the caret opens them), each with its state (Planned, In progress, Put off). You can mark a step done or dismiss it there. A plan with nothing left closes by itself; one left alone for a week closes too, unless it holds a card you put off.
4. A CARD YOU DIDN'T ANSWER: if you type something else instead of tapping a card, it stays open until you decide it, or until a newer copy replaces it, and you can still decide it; deciding late does exactly what the card showed, no new turn.
5. WHAT STILL CLOSES: cards for deleting, payments, public shares, raising permissions (consent), launches and calls close when you move on, because what they would do could be stale.

Try asking

• Can you come back to this email later?
• What happens to a card I don't answer?
• Do you remember what we said to do later?
• Skip this one for now and remind me after the rest
• Where do I see your plan for this?

Questions I asked you

How to

1. TO ANSWER IT: just reply whenever you like, in the same chat — on the web or your phone.
2. TO DROP IT: say "drop it", "never mind" or "forget the demo question" and I'll never ask it again.
3. WHY I DON'T ASK AGAIN: I ask once. If the job needs the answer, I pick what fits best from what I know, do it, and say what I picked in the same message, so you can correct it with one word.
4. IN A GAME (trivia, a quiz, a riddle): before I ask each question I fix its answer where only I can see it, and I mark yours against that, so a guess is never marked right just because it sounds close.

Try asking

• Why do you keep coming back to the demo question?
• Drop that question
• Never mind what you asked earlier
• What did you ask me earlier that I didn't answer?
• Stop asking me about the demo

Your first look

How to

1. Tap the link at the end of my first message after you connected your email, or at the end of your "You're in": it opens your Home in the app. On a computer, Home is the column beside our chat; on a phone, it's first in the switch at the top. If that browser isn't signed in yet, it asks you to sign in first and then takes you there.
2. Waiting on you lists up to three people from your own mail and what each wants. Tap Draft next to one and the reply opens in our chat with them quoted: send it and I'll write the draft for your OK. Anyone you've answered since drops off the list.
3. Worth an intro shows only introductions I've already proposed; tap Ask to open the card and decide. Coming up is your next meeting from your calendar.
4. A first goal is the goal you told me in setup, or the one I think you're after. Tap Make it a goal and it's added to your goals, the same as adding it yourself on Goals. If it's already one of your goals, it says so.
5. On the waitlist, until your setup is done, your Home says "Link your phone to start onboarding." with Text me before your phone is linked, and "Pick up where we left off in our chat." with Open our chat once it is; either takes you to our chat. Once setup is done, your Home says Ready when you're in: what you picked and what I'll do with each the day you're in, the two ways to get in sooner, and See the price, which opens the same checkout as the payment link in our chat, with the free trial when you can have one. Nothing new is read from your email or kept for it.

Try asking

• What's my first look?
• What's the first look link at the end of your text?
• What does Make it a goal on my first look do?
• Why does my Home say Ready when you're in?

Side panel

How to

1. Beside a path or a campaign, just type in the panel: the About chip on its box means your message is about that one. Tap ✕ on the chip to talk about anything else; it comes back when you open another one. On the Goals tab, a goal's own chat sits in the middle for just that goal.
2. What we said about this, on a path or a campaign, and Earlier in Chat about this goal, over a goal's chat, jump to the newest message about that one; scroll up for everything before it.
3. Close the panel with ✕ in its header (or \ when you're not typing); the orb at the top right brings it back, and it stays closed beside Network until you open it again.
4. On a phone, Chat in the switch at the top (Home | Chat | Goals) opens the same conversation, and the orb opens it over what you're looking at, with its About chip.
5. A reply from someone in one of your campaigns comes back in the conversation with my drafted answer and a Send reply button, tagged with the campaign. The same draft waits on your phone and on the campaign: whichever you answer first is the one that counts.

Try asking

• Does my goal have its own chat?
• What is the About chip?
• How do I talk about something else beside a campaign?
• What does What we said about this do?
• Is the side panel the same conversation as the main chat?

Home: what's waiting on you

How to

1. On a computer, Home is the column to the left of our chat; on a phone, tap Home in the switch at the top (Home, Chat, Goals), and Chat takes you back. The most urgent stack comes first: what needs you, with things that move your goals higher, then what's worth knowing: opportunities I found, what I finished for you today, and news about people you follow.
2. The Now card at the top shows the time in your time zone, the weather where you are (from MET Norway) and your next meeting. The weather is for the city saved on your profile. With none saved, it uses the city your connection comes from, says “approximate location”, and never saves it. Tap the temperature to switch between °F and °C; it stays that way on your phone and computer.
3. The bar under it filters Home: All, Needs you, Goals, People or News, each with how many there are. Your choice stays on that device until you change it.
4. Cards that belong together stack under a tag that says what they're for: a goal and its path, a person, a company, or a kind of thing. The top card shows whole and the next two peek out under it; tap Show all (or a card behind) to open the stack, and Show less to fold it. A goal's tag opens that goal.
5. The short line under each card says why it's on your Home: how long it has waited or when it's due, then who it's for or the goal it moves.
6. Tap a card for its context and the button that opens it (the thread, the card, the task). Tap outside it or press Escape to put it back.
7. Tap Reply under it: the item sits as a quote above where you type in our chat. Tell me what you want (“pay it with the Mercury card”, “say yes for Thursday”) and I do it right there; the card reads as in progress until it's done, and if it isn't done within a day it comes back. The × on the quote cancels the reply.
8. Draft ready means I already wrote the reply. Tap it to read the whole email: who it goes to, the subject, every word and any file on it. The Send button names who it goes to and sends it in about ten minutes; Undo, on the line at the top of Home, stops it until then. Edit opens the words in place, and Send then sends yours.
9. An email I drafted in Mail that you haven't sent shows here whole, with everyone it goes to. Send sends it right away, as Send in Mail does; Edit changes the words (the subject stays the thread's); Discard throws the draft away, so it leaves Home and Mail. If it changed after you looked, nothing goes and you see it as it is now.
10. Not now takes an item off your Home until it matters again: when something about it changes, or it gets more urgent. There's no timer. It stays wherever it lives (your inbox, Tasks).
11. Never… takes it off for good, and I stop suggesting that person or company. You can say why in your own words (“we already work together”); I go by it and keep it to myself. Undo, straight after, takes it all back.
12. When more drafted replies are ready than the few on top, a line says how many. Handle them for me opens them all, each whole; one that does more than its words (a file, a meeting change) waits unticked for your look. The button says how many it sends; they go out in about ten minutes, and Cancel all stops them until then.
13. On the web, what I tell you on my own (a heads-up, a check-in) sits on Home under From xhuman for three days instead of in our chat; tap it to open the conversation it belongs to.
14. On a computer, the round button at the top left of our chat (⌘\ or Ctrl+\) hides Home: the chat moves to the middle and a big clock takes its place. News from me while it's hidden shows as a dot and one line under the clock; tap the line to bring Home back. That computer remembers it; your phone keeps Home as it is.
15. When my daily check thinks one of your to-dos looks off (a copy, a date that passed, one that should close or rest), it's a card under Needs a look that says why. Tap it: its button does the one thing the check suggests (Mark it done, Drop the copy, Park it until a date), and you can take it back right after. If the to-do moved since the check looked, the button no longer applies and nothing happens.

Try asking

• What's waiting on me?
• Why is this on my Home?
• How do I get rid of something on my Home screen?
• What does Reply on a Home item do?
• Why are these grouped together on Home?
• What does the stack on the left of the chat show?
• What does the tag over a stack on Home mean?
• Can I filter my Home to just what needs me?
• Why does my Home say approximate location?
• Can I send a drafted reply from Home?
• What's the difference between Not now and Never on Home?
• What's the line on Home saying more replies are ready?
• How do I hide Home?

Goals

How to

1. Choose Goals in the switch at the top (Chat | Goals; on a phone, Home | Chat | Goals). Each goal is a bare title (New goal first; then any campaign of yours that's running on its own). Goals I think you have wait under Suggested: ✓ to have me work on it, ✕ if it's not a goal.
2. A goal's chat is in the middle: type to steer it. While a path or a campaign is beside it, the box shows an About chip naming it (✕ on the chip for the whole thing). Earlier in Chat, at the top, jumps to the main chat.
3. Plan, in the bar over the chat, shows the plan beside the chat; the orange count ("2 need you") goes to the first path that needs you. Each path is one row: who's on it, its name, one line, and a dot for its state. Choose a row for the whole path; Back returns to the plan, and ✕ closes it.
4. A running campaign that serves a goal is listed under that goal as one of its paths, never as a goal of its own; a campaign no goal fits stays a goal by itself, and a goal chosen for it by hand wins.
5. All campaigns, pinned at the foot of the Goals list, lists every campaign you started, filed or not (running, ready, paused, finished, being set up); a filed one is in its plan too. Choose one for its page beside the chat: People (Table or Pipeline, why each one matches, who can introduce you), Emails (each step's template, which you can edit) and Replies and activity. Open the table, on a campaign's path, does the same.
6. Paths ready to send show Send notes (it says how many), which sends the notes I drafted in your name. The first ones go out at once, within your working hours and today's limit; Undo stops the ones that haven't left, and a note that has gone out can't be taken back. A path I suggest that isn't drafted yet has Start instead.
7. Re-plan, at the bottom of the plan, has me rewrite it now from how things stand, a few times an hour at most; it shows only on a goal with a written plan, never on an always-on mission, a goal I only guessed or a finished one. Not now shows only on one of my ideas that isn't under way yet, and sets it aside for a week; then it comes back by itself. A to-do on your list that waits on you with no conversation of its own has its own Done right on its line (Undo for 30 seconds after).
8. Pause, at the top of the plan, stops everything under it until you press Resume; Pause on a path stops just that path. If I started one you don't want, press ••• beside Pause and then Not a goal.
9. Each row of the Goals list has an × (a long press on a phone), and Select in ••• picks several: × removes a goal or archives a campaign (it stops, and is kept, marked Archived); the toast restores either. Delete for good is inside it.
10. Activity, the first tab, shows what I'm on this minute for that goal (Now, a spinner while it runs), then Coming up (my next look and its time, what's left), then what's finished, by day in your own time zone.
11. History, at the bottom of the plan, is that goal's own record of the last month: every task done or stopped for it and its paths in the past 30 days, newest first, with its result or why it stopped.
12. Stop everything, at the foot of the Goals list, pauses everything I do on my own; the line across the top then says what stopped, with Restart. In What's new with them, each person I follow has Check now, Stop and Delete, and under Pages and news I check, each news alert, page watch and check I wrote has Delete (a page watch also has Check now). Delete removes it on the tap.
13. How goals work, at the foot of the Goals list, explains it all, and its What xhuman knows about you shows who I think you are, your Offering and the Background agent switch (on or off). New goal starts one in your own words, its chat ready. No "Done when" yet? Press Add a finish line in the plan and finish the sentence in its chat.
14. Always on holds what I keep up for everyone (replies you owe, the people who matter, the places you'll be). What's new with them, under Always on, lists what changed for everyone I follow, and Watch someone adds one. In the background, under a line below your goals, holds What I'm doing for you and Automations: everything I do on a schedule or when something happens.

Try asking

• How do I talk to you about one of my goals?
• What paths are you taking for my sales goal?
• How do I start a path you suggested?
• How do I pause a goal?
• How do I set what done means for a goal?
• Who is on one of the paths for my goal?
• What does Not a goal do?
• Where do I see my campaigns?
• Where is my campaign's page?
• Where's the plan for my goal?
• How do I get you to rewrite the plan for a goal?
• How do I set aside one of your ideas?
• How do I see the whole table for a campaign?
• Where do I ask about just one goal?

Tasks

How to

1. Open Goals (the button at the top of the chat or the Goals row in the sidebar, or /tasks). It opens on the Goals tab; how a goal's chat, paths, plan and campaigns work is in the Goals entry, so I point you there instead of repeating it.
2. The luck my work made for a goal (people found and reached, replies, meetings, intros, deals) is at the top of that goal's page, this week or since the start.
3. The old surfaces now live inside Goals, and their old links open there: the Engine and Running (/tasks) and Chasing (/tasks) are the Goals tab, Watching is (/tasks?view=watching), Automations and scheduled sends are (/tasks?view=automations, in its "Emails going out", /tasks?view=automations), and campaigns are (/tasks/campaigns); standing rules are in Settings › How it works for you (/settings/how-it-works?section=habits).
4. Every move you make goes to the place it lives — pausing a campaign pauses the campaign. Only a drafted reply (shown to you whole first), a delete for good (except the Watching page's Delete, which removes it on the tap), and a send that skips an "only if" you set ask you once.
5. To see one campaign's people, open its page from the Goals tab: People, Emails, and Replies and activity, with the chat beside it.
6. To stop everything I have running on my own, press Stop everything at the foot of your goals list, or tell me in chat. While it's stopped, a banner at the top of the Goals tab has Restart (so does Settings › How it works for you), and it shows what will send before anything goes. On the Watching page a person or group I follow can be checked now, stopped (Follow again brings it back) or deleted, a page watch can be checked now or deleted, and a news alert or script check can be deleted; and I can list everything running right now and show you the ones that are done.
7. Needs you is a decision for you to make, or a date close at hand: a reply you owe, an approval, an answer, a confirm, or a promise or to-do due within two days or overdue. A to-do with no date, or one due later, is on your to-do list, never counted in Needs you. A campaign's people waiting for you to check them, its people with no email yet and its emails with no copy are "Have a look" rows on its path, never counted in Needs you either.
8. To get my plan for something, ask me for it, or open the plan panel on a goal.
9. Tell me to add something — “add follow up with Nathan Bell as a to-do”, “remind me to pay the invoice Friday” — and it goes on the list right away. Your own rows are never filtered out or ranked away; they stay until you mark them done.
10. Ask me to mark a to-do done, snooze it until Monday, or dismiss it; every change can be undone for 30 seconds. If my daily check marked a to-do of yours done and you want it back, ask and I undo it.
11. Tell me to close a promise and I close it from chat, even one with no row in Tasks, and your promises agree. A promise with no date named is never called due or overdue.
12. I split tasks into steps: a task with parts gets sub-tasks, and a step says who will do it — you, me or both; saying so starts nothing. To have me do a task on my own, tell me to take it. You can also tell me to take a chat's whole plan and finish it on my own; I carry it on in that chat.
13. A promise you gave me that I finish is closed as my word, and you can ask me to reopen it. Questions I come up with myself about a goal are offers you can answer or skip; a question a job of mine is stuck on needs you.
14. Whoever emails you and fits a goal is a lead, filed under that goal; it is not a task. The task is the action I wrote toward them, like replying to offer a demo.
15. Work already under way that nothing ties to a goal yet I file under the goal it serves, marked My guess; say it's not really that goal and I send it back and never put it there again.

Try asking

• Where do I see everything you're doing?
• What needs me?
• Show me my campaigns
• Who replied in warm reconnect?
• Where are my rules?
• Add follow up with Nathan Bell as a to-do
• Remind me to send the deck tomorrow morning
• Don't let me forget to send the deck
• What should I do next?
• Keep track of this for me
• Open task T-42
• Mark the deck promise to Ana done
• Close those two promises, they were tests
• Split this task into steps
• What are you working on right now?
• Show me what is running
• Why isn't my to-do in Needs you?
• What is xhuman doing for my goals?
• Which intros did you make this week?
• Can I decide this right from Tasks?

Replies you owe

How to

1. WHERE: open the Tasks page and look under Needs you for “Replies you owe”. Each row is one conversation waiting on you.
2. HOW A ROW CLOSES: you reply to it from any of your mailboxes (the row closes once your reply has gone out), or you mark it no reply needed (Done or Dismiss on the row). Remind me parks it until later. If they write again, the old row closes and a fresh one takes its place.
3. WHY AN EMAIL IS STILL THERE: nothing has answered it yet — a row only closes when you reply or decide, so one you handled outside xhuman closes the next time I see your sent reply. Archiving or snoozing it in your own mail app only parks it until the conversation comes back to your Inbox.

Try asking

• Where do I see the emails I still need to answer?
• What's Replies you owe on my Tasks page?
• Why is that email still on my list?
• How do I get an email off Replies you owe?

Daily to-do check

How to

1. WHAT I CHANGE: at most 3 of your own to-dos a day (marking one done, dropping a duplicate, parking one, fixing a date or the next step), and only while the fix still fits the to-do as it is. Everything else I find waits for you.
2. THE ONE MESSAGE: one message, on the chat you write in most (email only when no chat is linked), naming each to-do I changed and what I did. Nothing new to say, no message.
3. PUTTING ONE BACK: tell me which one ("put the tripod one back") and I reopen it if I marked it done, dropped, parked or rested it; if I moved its date, change it back in the to-do's date editor or tell me the date you want. What I tidied also shows once on the Goals tab's Watching and Automations pages with Undo.
4. NOTES ONLY: anything from an email, a meeting or another person, and anything with no fitting fix, is its own card on Home under Needs a look; a card with a fix is one tap, and a note opens where the to-do lives.
5. Say "look over my to-dos" and I go through them now (no message then: you're looking).

Try asking

• What's the daily to-do check?
• Why did xhuman change my to-do?
• How do I undo what xhuman tidied?

Heartbeat

How to

1. WHERE: Goals → Automations, the first row, "Heartbeat". The switch pauses it (it won't look again until you turn it back on); the menu beside it sets how often it looks, from every 15 minutes to every 4 hours.
2. WHAT IT COSTS YOU: the quick looks are on us. When one finds something worth a closer look, that look comes out of your agent share, never credits you bought; as the share runs low it looks less often, so the rest of my background work keeps going.
3. WHAT IT DOES NOW: when something needs you (a birthday tomorrow, a reply someone is owed, a to-do that slipped), I take a closer look and text you once, in the chat you use most, with a draft that waits for your yes when one helps. Usually it stays quiet, and it counts toward the few texts I send you on my own each day.
4. TELL IT WHAT TO KEEP IN MIND: say it in plain words ("watch Ana's reply") and I add it to the notes the next check reads.
5. Stop everything pauses it too, and it waits while I'm already working for you.

Try asking

• What's the heartbeat?
• How often does xhuman check on my day?
• How do I pause the heartbeat?

What I did, history

How to

1. OPEN IT: on the Goals tab each goal keeps its own History, in its plan: what got done or stopped for that goal and its paths, newest first. For everything at once, ask me.
2. ASK ME: I can read the same list, so "what did you do today?" or "what did I approve this week?" gets an answer from it.
3. STOP SOMETHING RUNNING: ask me and I can stop a command on your computer, a background job or one of its items, a timer or a watch. Stopping one only ends that one thing. Stop everything is a separate button at the foot of your goals list on the Goals tab.
4. BACKGROUND AND UNASKED: background checks I ran on my own, and messages I sent you that you didn't ask for, are rows too; a message's row names each channel (Telegram, text, WhatsApp, email) and whether it was delivered or didn't arrive. Some background work isn't listed yet, and the page says which.
5. WHAT IT CAN'T DO: it doesn't cancel or unsend anything that already went out. For older approvals it doesn't record where you answered (which chat or device), so those rows don't say.

Try asking

• What did you do today?
• What did I approve this week?
• Is anything running right now? Stop it
• Show me everything you've sent for me
• Where's the history of what you've done?

Hand a task to xhuman

How to

1. On Tasks, open a to-do's owner picker and choose “Give it to xhuman” — or press “Give it to xhuman” in New task. Its row then says what I'm doing: working on it, checking back at a time, needs you, reported back, or couldn't run (with Run again).
2. Or just ask me in chat: “take care of this for me”, “research flights to Lisbon tomorrow at 9 and tell me” — I write it down as your task and start on it now or at that time.
3. Say how you want the result — “…and email me”, “…text me” — and that's where it goes, and no other channel (by email if I can't reach you there), even if you're in the app when it finishes. Otherwise it reaches every channel you've linked. Change it any time (“actually, email me”) and it reports there from then on, Run again included.
4. What your words didn't ask for — a send to someone else, a delete — waits for you on its card, nested under the task, and so does a payment's one price question. A draft printed whole on your phone can be approved there. Once you decide it, I carry on and finish the task.
5. A promise you made someone can be given to me the same way, from its owner picker: I draft what keeps it and ask you before any email goes out — the draft waits for your yes under the promise — and the email to the person you promised is what closes it. If I finish it some other way, I close it as my word, and Reopen brings it back.
6. A plan I keep for one of our chats can be given to me too: press “Give it to xhuman” on the plan's row and I carry it on in that same chat on my own, ticking its steps as I go. Its steps move with the plan, so one step can't be given away by itself.
7. To stop me, choose “Take it back” in the owner picker, or just tell me in chat (“take it back, I'll do it”); it stays on your list as yours, and nothing more starts on it (a step I'm already in the middle of finishes). Mark it done or dismiss it, and I stop too.

Try asking

• Can you just do this task for me?
• Give this to xhuman
• Research this tomorrow morning and tell me what you find
• What happens when I hand you a task?
• How do I take a task back from you?
• Can you take care of something I promised someone?

Chat history

How to

1. Ask “what chats do I have?” and I list them newest first.
2. Ask me to delete specific chats, or all of them — I delete exactly those and tell you what went (never the chat we're in).
3. Deleting is permanent: the chat and its messages are gone, with no restore.
4. Two chats at once: the orb button at the top right (or \ when you're not typing) opens a side chat — a separate conversation beside the one you're in. Its clock button reopens a past chat there.
5. Jump to any moment of a long conversation (past eight back-and-forths): point at its left edge, at the column of short lines, one for each message of yours and my reply. A card shows when it was and what was said; click to go to that moment, older messages loading if they must. On a phone, scroll and the lines show on the right edge: drag along them and let go. With a keyboard, Tab to them, then ↑ ↓ and Enter.
6. Reply to, copy or rate one message: open its menu with the ⋯ that shows beside it (or a long press on a phone). Reply puts a one-line quote of that message above the message box (✕ drops it), and what you send next answers exactly that message — one of mine or one of yours — with the quote on your bubble. Copy copies the message's text, links included. Thumbs up or thumbs down on one of my replies tells xhuman how it did; tap the same thumb again to take it back.
7. Talk instead of typing: the mic in the message box dictates. Tap it and speak — your words appear in the box as you talk; ✓ (or Enter) keeps them to edit and send, ✕ (or Esc) throws them away. Or hold the mic, talk, and let go to keep. Nothing sends on its own. The mic is speech to text, not a voice call.

Try asking

• Delete all my chats
• What past conversations do I have?
• Clear out every chat from before last month
• Can I have two chats open at the same time?
• Open a second chat next to this one
• Can I dictate instead of typing?
• Can I reply to a specific message?
• How do I reply to one of your messages?
• How do I copy one of your messages?
• Where did my earlier chats go?
• How do I see an old conversation?
• How do I jump back to an earlier point in this chat?
• What does the orb do?

The chat box

How to

1. The + in the chat box attaches a file, and that's all it does: tap it to pick a screenshot, voice memo, PDF, Word doc, PowerPoint deck, CSV or spreadsheet (on a phone you can also take a photo), or drop or paste one in. A voice memo is transcribed the moment it's attached, its words show on the file so you can check them, and when you send, those words are your message — the same as sending me a voice memo by text. What the chat is about shows as small chips just above the box.
2. Add objective left the chat box: your goals live in the Goals tab, each with its own chat and a chat for each of its paths, so open the one you want to talk about there (New goal there starts one). Or just tell me what this chat should work on.
3. There's no search switch in the chat box any more: a people search keeps looking on its own when it comes up short. To search only a teammate's network, just say so: “search Ana's network for designers”.
4. The suggested replies under my latest reply are gone, and so are the suggested next actions. Just tell me what you want next, in your own words.
5. Which model answers you is set in Settings › Model, not in the chat box. Pick one there and your next message uses it.
6. A campaign's settings — how it sends (Approve each email, Approve replies or Just run it), whether it keeps finding new people, its pace (up to so many a day) and the mailbox it sends from — are changed by telling me, like “send up to 50 a day”, “stop finding new people”, “send from my other mailbox” or “just send them”. Turning a campaign up to Approve replies or Just run it is yours alone: I show you a card and it changes only when you say yes.

Try asking

• Where did the model picker go?
• How do I change which model answers me?
• How do I attach a file in the chat box?
• Where did the campaign settings under the chat box go?
• Where did the suggested next actions go?
• Where did Add objective go?

Search your chats

How to

1. Press ⌘K (Ctrl+K on Windows), or click the magnifier at the top right. On a phone it's Search in the ⋯ menu at the top.
2. Type the words you remember. Every word has to be in the message, the last one can be half typed, and capitals and accents don't matter.
3. Results appear as you type: the best match first as the Top hit, then every other message that matches, each as a short bubble of what was said with your words picked out, the day it was said, and the chat it's in when that isn't your main chat.
4. Click one, or move with ↑ ↓ and press Enter: its chat opens at that message, which lights up for a moment. A message from your main chat (or from an earlier conversation, which is part of it now) opens in the main chat; one from a goal's chat opens that goal (or that path) in the Goals tab, with its chat beside it.
5. It matches words, not meaning, for now: search for words the message actually used.

Try asking

• How do I search my chats?
• Can I search my old conversations for a message?
• Where's the search for past messages?
• Find something you told me in another chat

Memory across chats

How to

1. It works on its own: ask about something from earlier in this chat, or from another chat this week, and I use what I kept — for example “what did we decide on Telegram this morning?”.
2. A summary keeps your own words and what I said and did — never what an email, a web page or a forwarded message said.
3. When you forget a memory, or remove or change a standing rule, the summaries that may have repeated it are cleared.
4. Deleting a chat deletes its summaries.
5. Only the last three days of your other chats, and at most three of them, are in view at once — as one line each.
6. When you point at something this chat doesn't hold — “my hotel”, “the one we picked”, “that search” — I look inside your other chats from the last two weeks before asking you to repeat it. Say how far back if it was longer ago, up to 90 days.
7. What you said in another chat is background here, not a request: I won't act on it, send or buy anything because of it unless you ask in this chat.

Try asking

• What did we discuss on Telegram this morning?
• Do you remember what we decided earlier in this chat?
• Do you know what I said in my other chats?
• Can you look at what I told you on my phone?
• What happened while I was away?

Inbox triage

How to

1. Ask “what do I need to respond to?” and I pull the mail awaiting your reply.
2. Say “help me clear my inbox” and I sort the last 30 days of every connected mailbox (ask for fewer — “just today”) into Needs you, Quick notes, No reply needed and Older; an overview card keeps the counts.
3. Needs you comes first, one card per conversation: the email plus an editable reply in your voice you can format, Cc and attach to — one offering times from your calendar too: the times you keep in it are still booked when they pick one (only the note confirming a meeting they booked sends exactly as written). A locked conversation's card comes with no draft — its text never reaches a model — and a sensitive one's only after your OK; you can still write your own. Send & next (⌘↵) sends it now; No reply needed (E, or say “skip”) clears it; Remind me (H, on the card — I can't set it from chat) takes a time in words — “tomorrow”, “next monday” — and brings it back then; R jumps to the rewrite chips (Shorter, Warmer), or tell me any other change in chat — “decline politely”, or your own text to use instead of the times.
4. Quick notes are for people who had the last word — “Thanks, Ann!”, or “Great, see you Friday.” to someone who booked a time with you. All of them on one approval card: approving sends the ones you tick now, all of them — they can't be unsent — and marks the ones you leave unticked no reply needed; “Done with these — no reply” clears them all instead. A booking acknowledgement is never sent automatically — it waits for your approval; only a campaign on Just run it, or a rule of yours that books meetings, confirms a meeting it booked itself.
5. No reply needed clears in one tap — “Clear N” on the overview, or say “clear them” — with an Undo. Older mail, past the window, is one decision: “review the older ones” walks the ones that may still need you, “clear the rest of the older ones” clears the others.
6. Every decision writes through — to /mail, the campaign's People table and your Worth doing list — not just this chat. A reminder also archives the conversation in Gmail until it wakes (Outlook keeps it in the Inbox); No reply needed leaves Gmail and Outlook as they are.
7. In a campaign chat, “inbox zero” walks that campaign's replies only — everyone who replied and hasn't had an answer, however long ago; “whole inbox” widens it to every connected mailbox.
8. Say “handle them for me” about the people waiting on an answer from you, and I answer them in one go: each drafted in your voice, in its own thread, from the mailbox it came to — Gmail or Outlook. One I can't send as it stands (a placeholder still in it, a file it needs, or the conversation moved on) stays with you, and I tell you which and why in one line. Your ask is the yes: every answer goes out right away, however many, and a sent one can't be unsent. If one of those conversations moves before they go — they write again, or you answer it yourself — nothing in the batch is sent.

Try asking

• What emails do I still need to answer?
• Help me get to inbox zero
• Let's get to inbox zero on this campaign
• Send the quick notes
• Clear the rest of the older ones
• Handle them for me

Inbox learning

How to

1. WHY A MESSAGE WAS IMPORTANT: the reply alert and the /mail row carry one line built from your own counts — how many messages like it you answered, out of how many, and how fast. It is never my guess: the counts are recomputed from your mail every few hours, nothing is learned in a way you can't check.
2. WHY I DIDN'T PING YOU: a message is kept quiet only when your history is clear — at least 8 messages of that kind with essentially none answered, or a stranger's 4th unanswered touch — and never when you've written to that person in the last 90 days, ever answered them, answer their company, are already in the thread, or the message is about something of yours specifically. A quiet message is still in /mail with its reason on the row, and it never reaches the daily brief.
3. Replies, forwards and meetings booked with the sender within a week all count as handled; mail you were only CC'd on, or one of many recipients of, doesn't count either way. Replies xhuman sent on its own never count as yours. Gmail and Outlook alike.
4. DRAFTS: up to three of your real replies to the most similar past messages are shown to the drafter as examples of HOW you answer (decline politely, propose times, redirect) — with every name, number, date and link masked, never as facts. If a draft still repeats a specific from one of them, it's rewritten without them.
5. A rule that lets me answer on my own still answers quiet messages it covers — only the alert is held back.

Try asking

• Why did you think this email was important?
• Why didn't you ping me about the email from Acme?
• Why is this email marked quiet?
• Why was this email quiet?
• Why do you think this email is important?
• You didn't ping me about that email — why?
• What made this email important?
• Does my inbox learn from what I usually answer?

Old drafts

How to

1. WHY A DRAFT CHANGED: it was a campaign reply offering times that passed or filled up since it was written, and nobody had edited it — so its times were re-checked against your calendar and it was redrafted, the same way asking me to refresh its times would. Only the offered times trigger this.
2. WHY A DRAFT SAYS IT'S OUT OF DATE: what it was written on moved — its offered times passed or filled up, a standing rule of yours it was written under changed, or it was already sent or set aside — and it could not be redrafted on its own: you edited it, or it isn't a campaign reply whose times can be refreshed in place (an inbox reply, a booking or meeting change, a follow-up), or the redraft didn't go through. It's kept exactly as it was. Change it yourself, ask me to rewrite it or offer new times, or skip it.
3. WHERE ONE WENT ("I never sent it"): what happened after it was written settled the email, so it was put away with its reason — you wrote back to them in that thread and they hadn't written since; a newer email of theirs has its own (that one is current); or later mail or a meeting answered it (you met; the meeting it asked about was booked; they withdrew it). The reason is kept on it and said where it was put away (the Inbox Zero walk; the campaign list names what it cleared). Nothing was sent or deleted — say so; to answer anyway, write from /mail or ask me. Age alone puts nothing away, and a meeting BEFORE their email is not evidence about it.
4. WHEN "OUT OF DATE" NAMES WHAT HAPPENED SINCE: an answer is still owed, but the text no longer fits — it offers times for a meeting since held, booked or cancelled, or they changed or added to their question. It's kept as written; rewrite it, or ask me to.
5. The send checks the offered times again at the click, as it always has: a reply never goes out offering a time that has passed or is taken. A draft that's out of date isn't offered as a one-tap send on your phone.
6. Inbox Zero cards, reply alerts, /mail's drafted replies and the campaign list are checked this way; inbox and campaign mail, Gmail and Outlook alike. What happened since is read every few minutes for ones written in the last two weeks; an older one is read again only when the Inbox Zero card or the campaign list is about to show it — elsewhere it shows what was last read. A draft whose send books, cancels or moves a meeting is never put away this way: at most it says it's out of date, and you decide.

Try asking

• Why did my draft change?
• Why does this draft say it's out of date?
• This draft offered times that already passed — why?
• Why was the reply to Dana redrafted?
• Why was that cleared — I never sent it

Heads-up

How to

1. HOW OFTEN: at most one message a day, never in your quiet hours; the next one after the night covers what came in while you slept. A heads-up that stops being true (the water is back on) isn't sent late.
2. WHERE IT COMES FROM: a heads-up is my summary of one received message. If you ask where it's from, the message is in your inbox and I can open it and tell you what else it said.
3. TO STOP THEM: tell me to stop the heads-ups and I turn that notification off for good (it's a toggle in your notification settings too, and quiet hours apply as usual). Telling me to turn them back on does the reverse.
4. WHAT I NEVER DO: a heads-up carries no links, no phone numbers and no addresses from the email, and I never act on anything an email asks. Mail you've already read isn't sent.

Try asking

• Why did you text me about the water being shut off?
• How often will you send me heads-ups?
• Stop the heads-ups
• Where did that heads-up come from?
• Can you tell me things from my email without me opening it?

LinkedIn messages

How to

1. Connect LinkedIn first — ask me and I'll show the connect card; live sync needs a paid plan (any paid tier qualifies).
2. Ask “did anyone message me on LinkedIn?” and I list your unread chats.
3. Ask me to draft a reply — I show a Send card you approve; I never send on my own.

Try asking

• Any LinkedIn messages I need to answer?
• Draft a reply to the recruiter who messaged me on LinkedIn

LinkedIn steps in campaigns

How to

1. Connect LinkedIn first — ask me and I'll show the connect card; sending needs a paid plan (any paid tier qualifies).
2. Ask for connection requests to a search — “send connection requests to founders of recruiting firms on my LinkedIn” — and the campaign opens with an invite (nothing else unless you say so), paced under the daily limits, launched when you say go.
3. Ask for a campaign to people in your network (“message everyone I know who went to CMU”) — once LinkedIn is connected, the sequence gets a connection request and a message after the opening emails by default.
4. Add or edit the connect / message steps in any sequence; launching the campaign is what starts sends, gradually, within your account's allowance.
5. The strategy card suggests a channel mix from the audience and says why — open with a connection request when most people have no email address, email first and then an invite when most have both. When a preset adds the invite and the message — or I add an invite with no note of yours — each person gets their own short note (under 200 characters) and their own message, written in your voice from your campaign copy and what we know about them, and checked before it goes: a note that can't be stood behind is dropped (the invite goes without one), a message that can't is skipped — your email is never pasted in. Text you write or dictate for the step goes out exactly as written.
6. The limits belong to the account, shared by every campaign: 15 invites a day and 80 a rolling week by default, in force from the first day — there is no warm-up period, and those are OUR numbers rather than limits the platform imposes (its own invite limit is weekly, about 100). Say the number you want — “15 invites a day”, “raise my invite limit to 25” — and I set it, up to 50 a day / 200 a week; or Settings → Email & calendar. A campaign's own cap (“cap this campaign at 10 a day”) only takes a share of that allowance.

Try asking

• Send connection requests to founders of recruiting firms on my LinkedIn
• Connect with every VP Sales at Series B startups
• Message everyone in my LinkedIn network who went to Carnegie Mellon
• Add a connection request before the first email in this campaign
• How many invites can I send a day?
• Change the number of invites per day

Default channels

How to

1. Only a NEW campaign, and only when you haven't named a channel for it — asking for email only, pasting your own copy or using a template all count as choosing. The campaign's strategy card says why it was added, with an Email only button beside it.
2. The shape: the opening emails, the connection request, a message once they accept, then the rest of the emails. Someone who never accepts only loses the message — the emails keep their days.
3. “Make this campaign email only” — one click on the card, or say it and I switch that campaign to email only. “Stop adding it to my campaigns” — I set the account to Email only for every new campaign (Settings → Email & calendar has the same switch). A campaign you already have never changes.
4. Invites go out gradually under your account's allowance — 15 a day and 80 a rolling week by default, our caps and not LinkedIn's — shared by every campaign. Who gets one first is decided at launch (people with no email, then your best fits; the rest after their emails), and nobody's emails wait for an invite. Each note and message is written for that person right before it would send.

Try asking

• Why did my campaign add LinkedIn?
• Why was LinkedIn added to my campaign?
• Make this campaign email only
• Stop adding LinkedIn to my campaigns
• Why are my invites waiting?

Calendar

How to

1. Read-only on a connected account is enforced by us, not the provider: every write on it is refused on our side, so it never reaches them; the access you granted Google, Microsoft or the app at connect is unchanged. For a calendar, ask — or say so when you connect it: I keep reading it and never send, draft or change anything in it.
2. Connect Google Calendar or Outlook if you haven't.
3. Open Calendar in the left rail: a week (or a day, or three) across every inbox; the Calendars list on the right narrows it to one inbox; the Calls page has the same week view under Upcoming.
4. Click any meeting for its card — join link, attendees (with any note they answered with), whether xhuman will join or skip it, Prepare me; your own events have Edit and Delete there. Once a meeting has happened, its card shows the notes' opening lines with Open notes (one click to the full notes) — or, when there are none, why (still being written, or the notetaker wasn't let in).
5. Every guest on the card shows the address they were invited at: click an address to copy it, the copy icon beside the guest count copies everyone's (never yours), and the envelope — Email guests — opens a new message with all of them in To, the meeting as the subject and the invite's own address as From. Nothing sends until you write it and press Send.
6. When your morning briefing wrote about a meeting (it covers today's and the next day's meetings), the card shows that part of it under Briefing — who you're meeting, how it came together, and the suggested angle set apart — with a link to the whole briefing.
7. An invite's card has Going? Yes / Maybe / No, Add a note (it goes to the organizer with your answer) and — on a repeating meeting — whether you mean this event or all of them. Suggest a new time picks a day and hours (your next free times are one click away) and sends it with a Maybe or a No: on Outlook as Outlook's own new-time proposal, on Google as a note on your answer, since Google's API has no proposal. It works on a meeting that already passed, too.
8. Click an empty slot (or drag across one) to create an event: title, calendar, time, guests, a Meet room on Google, place, notes — the button says exactly who gets an invite before it sends.
9. Ask what's on your calendar, or tell me to set up a meeting and I'll create it on your calendar, guests included, with no card to approve.
10. Ask for the follow-ups from your meetings over a period — what you promise and who owes you, from that period's calendar and mail, one line per person with the next step.

Try asking

• Follow-ups from my meetings
• What's on my calendar tomorrow?
• Where can I see my whole week?
• Show me just my work calendar
• Set up a 30-min call with Alex next Tuesday afternoon

Calls & meeting prep

How to

1. Open Calls in the left rail: Upcoming lists the next seven days with the notetaker's plan for each meeting; Past lists every recording with its notes.
2. Click Prepare on an upcoming meeting (or ask me to prep you) — I read the invite, your history with each attendee (even when they write from a new address) and the last call in the series, research anyone you barely know and their company, and write the brief here.
3. Ask me to join a meeting (paste the link), turn on auto-join for your calendar, or record one specific meeting from its row.
4. After the call, ask for the notes, action items, decisions, or exactly what was said.

Try asking

• Help me prepare for my meeting with Oxon Capital tomorrow
• Prep me for my 2pm
• What's on my calendar this week and which calls will xhuman join?
• What meetings do I have tomorrow?
• Join my 2pm Zoom and take notes
• Summarize what we decided in this morning's meeting

Meeting recordings

How to

1. Ask “find the meeting where we discussed pricing” or “my meetings with Keegan last month” — I search the titles, notes and transcripts and quote the moment it came up.
2. Ask for a meeting's notes, decisions or action items, or exactly what someone said — I read the notes, or the transcript with its timestamps.
3. Ask “who was actually there?” — I tell you who the notetaker saw (or who spoke), apart from who was invited, and say so when it's unknown.

Try asking

• Find the meeting where we talked about pricing
• Who was actually at my meeting with Dana?
• What exactly did Keegan say about the timeline?

Call invites

How to

1. Only YOUR words count (recordings, Granola, or — where your commitments are tracked — your answer when I ask about a missed one). When the other side promises one instead, I check your calendar a day later and tell you if nothing came.
2. It goes out at once, however many guests — after your one tap too. Busy then, anything unclear, a guest who wasn't there, or a repeating meeting → nothing goes out and you get a one-tap Send. A move or cancel happens only with exactly one matching meeting.
3. The meeting's page shows what happened: sent, going out, waiting for you, or held and why.
4. Open Settings › Meetings › Notetaker for the calendar invite switch when I promise one: automatic (default), ask first, or off.

Try asking

• If I promise an invite on a recorded call, will you handle it?
• Did the invite I promised Jane go out?
• Stop sending invites I promise in meetings
• Stop sending invites from my calls
• Ask me first before those invites go

Late guests

How to

1. Only meetings YOU organize, with your notetaker in the room and you in it too; a late guest in someone else's meeting is never chased.
2. If anyone in the room can't be matched to an invitee (a phone named “iPhone”, a renamed device, someone who wasn't invited), nothing is sent — they might be the person who looks late.
3. At most 3 late people per meeting, emailed once, 6 to 15 minutes after the start, never for big meetings (more than 15 invited). It sends on its own and can't be unsent.
4. The meeting's page shows who was emailed and when. Settings › Meetings › “When someone's late”: on (default) or off.

Try asking

• If a guest is late, will you email them the join link?
• Stop emailing people who haven't joined my meetings
• Did you email Ana when she didn't show up?

Scheduling links

How to

1. Ask me to create a booking link with your availability and meeting length.
2. Send the link; bookings land on your calendar automatically.
3. To change one thing — hours, booking window, notice, duration — just say it; everything else on the link stays as it is.
4. When a campaign prospect asks for your link (or would rather not share an email for a calendar invite), the campaign sends them your link for a meeting that long — a personal one, so when they book it lands on their record, the follow-up stops, and your booking notice names the campaign and the person. Pausing or deleting a link tells you how many prospects hold it.
5. On Calendar your booking links sit under Calendars in the right rail (on a phone, behind the button at the top): Copy puts one's URL on your clipboard in one click; point at one and the hours it offers are shaded on your week, busy time on every connected calendar already taken out; click it to edit it beside the week, and the shading follows as you type; New makes one from your usual hours, live at once. A meeting booked through one carries a small mark.
6. Share times: pick up to six open times on your week, then Copy times (each with a one-click URL that books it) or copy the one-time URL instead — a single address offering exactly those times that works once: one booking, then it closes, and it expires when the times have passed. What you shared, and who booked, is listed under Shared times.

Try asking

• Make me a 30-minute intro-call booking link
• Set up a scheduling page for office hours on Fridays
• Let people book only 10 days out on my intro link
• Does my campaign send my calendly if someone asks?
• How do I share a few times with someone?

Trusted people

How to

1. Say “add Maya as a trusted person” or “connect me with Maya” and give me her phone, email or handle (or pick her from your network). By handle, she gets one message from her own agent and answers it like any text — a 👍, “sure”, a button, or the link.
2. By phone or email I check whether she's on xhuman (still on the waitlist counts). If she is, her own xhuman texts her one message with your profile, and a 👍, a yes or one tap connects you — you hear “Sent. I'll tell you when they answer,” with nothing to forward. If she isn't, I give you one text to forward from your phone (xhuman never texts people who aren't on it); its link shows your profile and one button, “Accept and join”, and that tap is her yes: she signs up and you're connected, with nothing more to answer (an emailed invite connects once her address is confirmed; a network swap you asked for in the same invite rides that yes). By email I send it from your own mailbox when you type her address yourself (over a plain green-bubble text, where anyone could fake your number, I check with you first); if I found the address myself, I ask you once before it goes. When she says yes you're connected; if someone else opens your link, I tell you who it is and ask you once.
3. You learn only whether a number or an address is on xhuman, never whose it is: her name and profile reach you only once she says yes. Someone whose “Who can ask to connect” is set to Nobody (or Only people I know, when they don't know you) looks just like someone who isn't on xhuman, and nothing reaches them from xhuman.
4. Requests show your name, handle and verification marks, never your phone number or email. The person you ask sees “phone verified” or “email verified”, and whether you're someone in their own contacts — never the number or address itself.
5. Connecting shares nothing by itself: her agent can ask yours, and I check with you before booking or sharing anything. Choose what each person gets, in your own words: “let Maya's agent book me directly”, “share my network with Maya”, “Maya can have my cell”, “stop telling me about every booking with Maya's agent”. Turn any of it off the same way.
6. I tell you when she accepts. If she says no or ignores it, you're never told — after two weeks the request just lapses.
7. When someone asks to connect with you, I bring it to your chat. Answer in your own words, or open the link to Accept, Decline or Block this person. If they also asked to swap networks, I ask both in one question, and one yes does both.
8. Network › Trusted (/network?tab=trusted) is one list of the people you trust, each with a line on what their assistant can do; people who asked you sit at the top and invites still waiting at the foot, with Resend. “+ Add someone” searches your network or takes a phone or email. Tap a person for switches of what their assistant can do, their circles, each conversation's transcript, “Stop trusting” and Block. Every person card in your network has “Trust” in its “⋯” menu, which starts the same add; on any person with an email, the Agents part of their Overview offers “Connect agents”, an invite you can email them (if they're already on xhuman, their own xhuman asks them instead).
9. Ask “who are my trusted people?” or “what can Maya see?”. “Disconnect Maya” ends it and stops everything you shared with her; “block Maya” also stops any future request, and she isn't told.

Try asking

• I'd like to connect a trusted person with me here — how do I do this?
• Add Maya Chen as a trusted person
• Connect my agent with Maya Chen
• How do I connect with someone's agent?
• Who are my trusted people?
• What can my trusted people see?
• Can your agent talk to Maya's agent?
• Let Bruno's agent book me directly
• Always check with me first with Maya's agent
• Bruno's agent can have my cell
• Share my network with a trusted person
• Block Bruno's agent
• Disconnect from Dana's agent
• Someone asked to connect their agent with mine — who is it?
• Will the person I connect with see my phone number in the connection request?
• Can someone find out if I'm on xhuman by my number when they invite me?

Circles

How to

1. Say “put Ana, Marco and Dad in a Family circle”. Anyone you know can go in a circle: a trusted person, or someone in your network who isn't connected with you — for them I offer a link to send, once.
2. Ask “who's in my Family circle?” or “what are my circles?”. “Take Marco out of Investors” is silent: Marco is never told, and nobody ever sees which circle you put them in — not their names, not who's in them, not that they exist.
3. Rules in your own words: “Family can see where I'll be”, “my co-workers' agents can have my booking link”, “Family's posts right away”. A circle's rules add to what you set for each person, and a “never” you set for one person always wins: “Family can see my calendar, but never Marco”.
4. Network › Trusted (/network?tab=trusted#circles): your circles are the chips above your trusted people — tap one to see who's in it, again (or Edit) for its rules and Remove, and “+ Circle” makes one. A person's own sheet shows their circles too. Everyone, Followers and Trusted people are always there, worked out from who follows and trusts you.
5. If you've tagged people in your network (say “poker”), I may ask once whether to make that a circle. Nothing moves without your yes.
6. Ask a circle: “who from my family will be in NYC in November?” or “ask my investors if they're free next Thursday”. I ask each person's agent on their own (at most 25 at a time) — nobody learns which circle they were asked as part of, or who else I asked. You get one answer as soon as the quick ones are in, then at most one more message when the rest land. For someone who isn't on xhuman I only look in your own email and calendar, and can give you a link to invite them.
7. “Where I'll be”: when your Family can see where you'll be, their agents can answer “will you be in NYC in November?” from the trips and places on your calendar (Google or Outlook) without asking you. Without it, your agent asks you first.

Try asking

• Put Ana, Marco and Dad in a Family circle
• Who's in my family circle?
• Take Marco out of Investors
• Can my family see where I'll be?
• Does Marco know I put them in Investors?
• Make a circle for my co-workers
• What circles do I have?
• Family can see my open times
• Rename my Investors circle to Angels
• Who from my family will be in NYC in November?
• Ask my investors if they're free next Thursday
• Does Marco know who else I asked?

Posts

How to

1. Say “post to my family: dinner at ours Saturday?”. Your own words go as you typed them, to your trusted people in that circle; nobody in it learns the circle's name. Someone in the circle who isn't connected with you yet gets nothing — I tell you how many. If you ask me to write it, I post what I wrote, and you can change it or pull it after.
2. A post goes out as soon as you ask — there's nothing to approve first. “Pull that post” stops every copy that hasn't gone out yet and takes down its reshares; anyone who already has it keeps it. “Change it to…” changes the copies that haven't gone out.
3. What comes in from people you trust waits for one evening note at 7 pm your time — the top three, who and what, and why it matters to you when that isn't obvious. On a day with nothing, nothing is sent.
4. “Tell me right away when Ana posts” sends Ana's posts as she posts them (up to 10 a day; past that they wait for the evening). “Stop sending me Leo's posts” mutes Leo — Leo is never told. A circle can say it too: “Family's posts right away”.
5. “Only Maya's launches right away” or “tell me right away when Leo posts about hiring” narrows an author to a topic, in your own words: when they post, I read it against what you asked for, and usually it isn't it, so it waits for the evening note like the rest (up to 10 right-away a day, shared with the people you named). “Stop telling me Leo's hiring posts right away” ends it, and the author is never told.
6. “Put the posts summary in my morning brief”, “weekly on Sunday”, “at 9 pm”, or “only when I ask” — then nothing is sent, and “anything from people I trust today?” reads it whenever you like.
7. When someone's agent asks yours something that can wait (“where will you be in November?”), it waits for your evening note instead of pinging you now.
8. “Follow Maya” (or Follow on her page) and her posts come in your evening note. She isn't pinged: new followers are one line at the end of her own note. “Post to my followers: we just shipped circles” reaches everyone who follows you — and your trusted people, who follow you by being connected; “post to everyone” also lets anyone read it.
9. “Who follows me?” shows only you who follows you, who's waiting for your OK, and who you follow — there's no public follower count. “Approve my followers first” makes new follows wait for your yes; removing a follower is silent.
10. Your page has three doors: Follow, Connect, and an in-person QR. “Only people I know can connect with me” (or nobody) sets who may ask; a stranger's request waits for your evening note, someone you know reaches you right away. “Show my code” fills your screen with a QR that works once for about ten minutes — they scan it, tap once, and you're connected. Nobody ever sees your phone number or email.
11. “Share my link”: your link is your page: xhuman.com, a slash, then your handle (like xhuman.com/ana). On your phone, Share my link at the top of your hub hands it to the phone's share sheet, so whoever gets it can follow you, connect, or ask your twin.
12. React to a post with a tapback (or a Telegram or WhatsApp reaction) and its author sees one more reaction — never a ping. Reply right after a post (“congrats! can my team get early access?”) and I pass it to its author, and only its author sees it.
13. “How did my post do?” or “who replied to my post?” — I tell you how many got it, opened it and reacted, and what people replied. Replies that ask you for something come in your evening note. Every post has its own page, linked at the end of the message, with the same numbers; breakdowns show only once a post reaches 10 people.
14. “Share Maya's post with my co-workers” or “send Maya's post to Dana” passes it on with Maya's name on it. Only a post Maya shared with everyone can be passed on; one she shared with a small group stays there. Nobody gets the same post twice: if Leo and two others pass it to you, it's one line saying so. If Maya pulls her post, every reshare goes with it, and “don't let people reshare this” turns it off for one of yours. “Who reshared my post?” names who passed it on.
15. “Like that” (or the heart on a post's page, or a ❤️ tapback) teaches me what you're into. Your likes stay yours until you say who may see them — “share my likes with my trusted people” lets their agents pass on what's relevant; “keep that one to me” keeps one like yours. Likes on a small group's post go only to its author. “Stop sharing my likes” or “mute Leo's likes” changes it, and nobody is told; unliking pulls a like back if no one has seen it yet. Likes are never a feed.

Try asking

• Post to my family: dinner at ours Saturday?
• Share with my trusted people: we're hiring a designer
• Pull that post
• Tell me right away when Ana posts
• Stop sending me Leo's posts
• Only tell me right away about Maya's launches
• Stop telling me Leo's hiring posts right away
• Anything from my trusted people today?
• Put the posts summary in my morning brief
• Only send me posts when I ask
• What did I post this week?
• Follow Maya
• Who follows me?
• Post to my followers: we just shipped circles
• Approve my followers first
• Only people I know can connect with me
• Show my code
• Share my link
• Send someone my link
• How did my post do?
• Who replied to my post?
• Reply to Ana's post: congrats!
• Did anyone react to my post?
• Share Maya's post with my co-workers
• Like that
• Stop sharing my likes
• Who reshared my post?

Talking to a trusted person's agent

How to

1. Ask in your own words: ‘set up coffee with Maya next week’ or ‘ask Maya what time works’. I reach out only when you ask me to.
2. Ask ‘what did you and Maya agree?’ and I will read you the thread.

Try asking

• Can you talk to Maya's agent?
• Set up coffee with Maya next week.
• What did you and Maya agree?
• Ask Maya if she is free Thursday.
• Talk to Maya about moving our call.

Daily briefing

How to

1. The setting is on by default at 7am your time; ask me to change the time or turn it off.
2. It only goes out once it's live on your account and a Gmail, Outlook or calendar is connected — ask me and I check, and say what's missing.
3. Ask “what's in my briefing today?” anytime.

Try asking

• What's my day look like?
• Turn off my daily briefing

Listen to your briefing

How to

1. Tap “🎧 Listen to your briefing” at the end of the phone message, or at the top of the email (sign in if it asks), then Tap to listen (allow the microphone).
2. Or just text or ask me “read me my briefing” and I send you the link.
3. The link in your phone message works for 24 hours; the email's opens through your xhuman sign-in, so a forwarded email can't call me as you. Hang up with the red button — whatever you asked for that wasn't done yet shows up in your chat, and one tap takes you back to it.

Try asking

• Read me my briefing
• Can I listen to my briefing?
• Read my brief out loud

Automations in your brief

How to

1. Nothing to set up: an automation scheduled near your briefing time (say a 6:55 or 7:00 morning digest with your briefing at 7) arrives inside the briefing's phone message.
2. Want it as its own message? Schedule the automation more than 15 minutes before your brief's hour, or after it has gone out.
3. Running an automation by hand, or one triggered by an email or a news story, always comes right away on its own.

Try asking

• Why did my digest automation come inside my briefing?
• Can my automation come with my daily brief?

Approve from your brief

How to

1. Nothing to set up: when you owe someone an answer and I've prepared one, your morning message shows it and asks, in a chat that proves it's you.
2. A yes (or a tap on its Send button) is all it takes.
3. Want it changed first? Just say what to change.

Try asking

• Why did my morning brief ask me a yes or no question?
• Can I approve an answer from my morning message?

Approve from your brief email

How to

1. Nothing to set up: when you owe someone a reply and I've prepared one, your brief email shows it.
2. Tap Send this reply in the email, check it on the page that opens, and tap the button there.
3. Want it changed first? Just tell me what to change.

Try asking

• Can I send that reply from my brief email?
• What does the Send this reply link in my brief email do?

Pre-meeting briefs

How to

1. It's on by default — open Settings › How it works for you, Before meetings under When it talks to you, to change how early (5 to 60 minutes), which meetings (only ones with people outside your company, or every meeting with other people), and where (Telegram, text, email, any mix).
2. Back-to-back? If the brief would land during another meeting, it goes out just before that one starts instead.
3. Switch it off for one meeting or a whole repeating series from that meeting's card on the Calendar, or just tell me.
4. Reply to a brief on Telegram or by text to ask about the people in it. “Send me one now” sends the brief for your next meeting.

Try asking

• Send me a brief right before my calls
• Text me 15 minutes before each meeting
• Email me the meeting prep too
• Stop sending me briefs before internal meetings
• Don't brief me for the weekly sync

Getting you ready for what's on your calendar

How to

1. It runs on its own for entries you made or accepted: I look into at most one a day, and may check that one once more nearer the time. Never during your quiet hours, never when you've muted me or set me to speak only when asked. All-day entries aren't looked at.
2. A flight, train, bus or ferry on your calendar gets one message from me: the evening before a morning departure, that morning for a later one. I write it from what I know — whether it looks booked (your email, and what we left unfinished), a reminder or a meeting that clashes with it, when to leave — and I offer the next step. If I can't write it, you still get the plain facts: when it leaves, check-in, when to set off.
3. It only looks things up and tells you — it doesn't book, pay or check you in on its own. Say yes to the offer, or ask me, if you want that done.
4. Tell me what you never want to hear about and I'll leave those alone, or tell me to stop reaching out on my own.

Try asking

• Will you tell me if I need to do anything before an appointment?
• Why did you message me about something on my calendar?
• Get me ready for what's on my calendar

Guest preps

How to

1. Settings → Meetings, Prep for your guests: the switch, and a preview of exactly what a guest gets.
2. Every meeting's card on the Calendar says who gets a prep and when — Skip this one, or Send one for a recurring meeting. Once sent, View shows the email as it left.
3. Only people outside your company, on one-off meetings of up to 15 people. It's written only from what the guest can already see: the invite, your public profile, and email you've exchanged with them — never your notes or briefings.
4. On a 1:1 you organized, the guest can pick a new time from your open hours and the invite moves (you're told); otherwise they reply to you. Anyone can opt out from the link at the bottom.

Try asking

• Does xhuman email the people I meet before our calls?
• Stop sending preps to my guests
• What do participants get before my call?
• Can guests reschedule from the prep?

Agent inbox

How to

1. Your agent's address is in Settings → Identity (copy it, or Save to contacts so it autocompletes). I can also just tell you — I know it.
2. To book a meeting with someone: CC my address on your message to them and say what you want ("adding my agent to find us 30 min next week"). I offer times from your booking link — you need an active one — and keep going until it's booked: the time they pick, a time they suggest if it's open, new times if none work. You're copied on every email; a question only you can answer, or moving a booked meeting, comes back to you. Only the people you CC'd me with — and anyone they copy in — can move that thread: anyone else who writes into it is handed back to you, unread.
3. To act on a message: forward it with a line on top. With no line, I summarize it and offer next steps as a numbered list — reply with a number. Your line counts only above the forwarded message's header (Forward adds one): if nothing marks where your line ends, I read all of it as the forwarded email — I summarize it, and nothing in it is taken as your instruction.
4. To file it: forward to your +save address. To have me take notes: add my address to the invite. When I need your OK, just reply and tell me.
5. When the browser signs up for a site with my address, that site's confirmation email — a code or a "confirm your email" link — is used by the browser for that sign-up only: it's accepted only from that site, checked by the mail server, while that task runs, and nothing in it is ever taken as an instruction.
6. Later mail to my address for an account I made for you — an e-ticket, a receipt, a password reset — is forwarded to your own email exactly as it arrived, attached, once the mail server confirms who sent it; it never makes me act, and I never read it. You can turn that off per account in Settings. Any other such mail stays unread.

Try asking

• What's your email address?
• Does my agent have its own inbox?
• Can I cc my agent to book a meeting?
• Can I forward my agent a message?
• Can I invite my agent to take notes?
• Can I reply to my morning briefing?

Voice calls

How to

1. Text me “can we talk?” or “let's talk through the Acme deal” — I reply with a link. Tap it, then Tap to talk (allow the microphone).
2. Just talk. Interrupt whenever you like. What I'm saying shows as a caption under the orb; pull the orb down (or tap Transcript) for the whole conversation, and type into it if you'd rather.
3. Hang up with the red button — the screen shows what we got done and one tap takes you back to our chat; whatever you asked for that isn't done yet, and anything I said I'd send you, is in our text thread a moment later. The link works for 24 hours and you can reconnect on it; each conversation runs up to an hour.

Try asking

• Can we talk?
• Can I talk to you instead of texting?
• Send me a link so we can talk it through
• Can I have a voice conversation with my agent?
• Can we talk about my week?

Call xhuman from the chat

How to

1. In the xhuman chat, tap the orb next to the microphone (the mic itself still just dictates into the box). Allow the microphone the first time.
2. Just talk — the orb grows into the centre of the page, with Mute, End and Type at the foot; tap Transcript (or pull the orb down) to read along. Back to the chat keeps us talking, and one tap on the orb opens the full page again.
3. Ask for anything: I work on it while we keep talking, and tell you what I found or did.
4. Tap End, wish me goodbye, or close the tab to hang up. Each conversation runs up to an hour.

Try asking

• Can I talk to you by voice in the app?
• How do I call xhuman from the chat?
• Can I have a voice conversation with you here?
• Can we talk instead of typing?

Phone calls

How to

1. Text or tell me “call Luigi's and book a table for 4 Friday at 7” (with their number) — I call and book exactly that.
2. Say what I may agree to (“anything Fri 7–8pm”) and any spending limit (“up to $40 for the deposit”), and I keep to exactly that; leave them out and I only ask questions and report back.
3. A standing rule or automation that asks for calls works the same way: if its words name a dollar figure (“every Monday, call the bakery and order the cake, up to $60”), each run of it may agree to spend up to that figure on the call, never more than $500.
4. When the call ends I message you what happened, with the transcript. If they offered something else (“they can do 8:30, not 8”), I ask once whether to take it, and your yes books it.

Try asking

• Can you call the restaurant and book a table for me?
• Call my dentist to move my appointment
• Can you wait on hold with the airline for me?

Sourcing & outreach

How to

1. Tell me who you're looking for and what for, in one message (“a VP of Sales who has sold software to banks, for my startup”). I don't ask whether you want them reached, why, or which goal it's for: I take that from what you said, and say in a line where I'm keeping it. I show who I'd look for on a card: the criteria as chips, one line on how I'll judge a fuzzy criterion like revenue (from news, filings and funding; unclear ones marked not sure, never guessed), where I'll look (people you can reach first, then the open web), and a few real people with how you'd reach each.
2. Shape it: tap More like this or Not like this on a person, ✕ on a chip, + Add, Change something, or just type it. The same card redraws with a new sample, and what changed is outlined. A preview costs almost nothing, and nothing is spent until Find.
3. Find starts as soon as you ask, with nothing to tap and no price in front of it: “Find ten…” looks for ten (a smaller first try only if you want one); with no number, it finds 10 first and offers to find more. It searches for exactly what the card shows, checks each person against each criterion, finds your warm paths, looks up their email addresses and writes each person's words — the email and, when LinkedIn runs, the connection note and message — all ahead of the sign-off. When it finishes, the message with what it found also gives the credits that finding them and their checks have used.
4. Then ONE sign-off card: the list first (why each person matches — tap one for each criterion's verdict, the fact found and where — who can introduce you, and the channel each gets; drop anyone with ✕, add a close match, or just tell me, like “skip LinkedIn for Marco”), how I'll reach them (the days and channels, and the honest daily pace), the emails (each step's template, the words everyone gets, edited in place; each person's version is on their page), and Start, the one tap, which names what it starts ("Start: 9 intros, 16 emails"): the first ones go out at once, and you can pause any time. Each email gets a personal line about that person, checked before it goes — in your saved voice, or, when you have none, in the voice of the email it sits in.
5. Ask about replies and status anytime; I can act across all your campaigns. A reply is any email from a campaign's prospect, in its thread or a new one they start: it stops their follow-ups and is handled like any reply — answered on Just run it, drafted for you below it. One they start about something unrelated to your outreach doesn't count: their follow-ups continue.
6. After launch, a campaign stays in the chat it was started in: its live card (status, emails sent, connection requests accepted, replied, meetings; Open and Pause), one message for each person who replies, with my draft under it, meetings booked, accepted connection requests that sent your message, and one daily line for opens, accepts and bounces.
7. Every campaign has a Status page in its module panel: whether it is live or paused and why (with the steps that restart it), a change log of what was done to it and when — by you, by me, or automatically — and the chats about it. Ask “is this running?”, “why did it stop?” or “what changed on this campaign?” and I read the same page.
8. A campaign can be paused (Pause, or “pause it”: resume it any time) or ended for good (End for good on its page, or “end it for good” / “cancel this campaign”): nothing it hadn't sent goes out, replies still come to you, and an ended campaign can't be resumed — start a new one to reach those people again. “Stop it” alone pauses.
9. A program I write may pause several people at once, and in the app that lands right away: the card reads “Applied N changes · Undo all”, and Undo all puts every one back, newest first — whatever couldn't be undone is named. Emails go out on your ask too.
10. Ask who opened or clicked — just say “turn on open tracking” and I'll switch it on for you (or flip it in Settings → Campaigns → Tracking), then I report opens per campaign and per person. Mail gateways and link scanners are filtered out, so the number is people rather than machinery. It only applies to sends made after it goes on, and image-blocking clients hide some real readers, so replies stay the hard signal.
11. Every people search also looks in your own network — your LinkedIn connections and the people you email or meet: the preview card shows the ones who match under “You know”, above the people new to you, and Find adds them to the list HELD: nothing goes to them until you say — email them yourself, or skip. The list tells you how many came from your network. (In onboarding, the “Where should xhuman look?” step sets the same thing.)
12. Tap a person on the preview card for their person sheet — the one contacts open in. A new face's sheet shows who they are, why they match the search's criteria (with the scorer's count, once scored), their latest roles with “Show all”, and one button that fits: “Add to list”, or “Draft intro” once you've chosen outreach; the second one and their LinkedIn are in “⋯”. Either button goes to me in the chat: I add them, or draft the intro for you to see first.
13. Contacts a teammate knows are marked “known by” with the name of whoever knows them, and are held the same way. Per person or in bulk: ask for an intro (ONE email from your inbox to that teammate bundling everyone they know on the list, drafted in your voice and approved like any send), email directly, or skip. Someone in both your network and a teammate's is your call, noted “also known by” with who else knows them. Nothing is skipped for you.
14. Every email address on a list is checked for deliverability the moment it is added or revealed, and the verdict stays on the person (valid, catch-all/unknown, invalid). Ask “were these emails verified?”, “check these addresses” or “verify the emails on this list” and I read the verdict on file or run a fresh check — dead addresses are dropped before they can bounce, and I say whose. A bounced address cannot be rescued by re-checking it; when an email bounces I look for a DIFFERENT address for that person in the background — your own contacts first, then a directory lookup — verify it, and ask you in the campaign chat before sending anything. Say yes to send there on the normal schedule, or skip. An automatic reply that they've moved — a new domain, a new address: your first email goes to their new address on its own, and I tell you; it goes out at once, within your working hours and today's limit, and removing them from People stops it if it hasn't gone yet. Not if you've already answered them yourself, or if that address isn't theirs, is yours, is on your do-not-contact list or won't take mail.
15. When a prospect replies pointing you to a colleague and gives their address, I verify it. On Just run it, if it is on their company's domain, ONE intro goes — your first email, with a line saying who suggested it — then your sequence as usual; otherwise it waits for you in People (Send intro). Nobody you removed is re-added, and my thank-you to them promises nothing. If they point to more than one person, nobody is contacted and I ask you who to write to.
16. Ask me to “add spintax”, “vary the wording” or “make each email a bit different” and I wrap short phrases in {wording one|wording two} — each recipient gets one of them, picked the same way every time for that person so what you preview is what sends. YOUR wording is always the first option, so nothing you approved is paraphrased away; edit any of them afterwards like ordinary text. The same button lives in Settings → Campaigns on a saved template. Every send records which wording that person got, so later you can ask which opener replies best.
17. One campaign can send SEVERAL sequences to different parts of its list, while it keeps running: “use this new copy for everyone we haven't emailed yet”, “give the investors a 4-email sequence and everyone else 3”, “put the rest on LinkedIn outreach starting Monday”, or split the whole list in one go (“these people get this, those get that, the rest get this later”). Each sequence has a name, its own steps and its own people; the Outreach panel's “Who gets which emails” shows every one with how it is doing. People already contacted always finish what they are on unless you say to move them — nothing already sent is ever re-sent.
18. A campaign's Campaign rules tab, top to bottom: Mode (the dial); New people (the On/Off switch, People to review, search depth, Run until); Limits (Most first emails a day — first contacts on a LinkedIn-first campaign — a hard cap, unlike Target first emails per day further down; total emails a day; days kept ready; monthly credit cap). New people: Off stops finding, and people already on the list are still contacted. Pause everything also holds sending until it is resumed. Stop finding & sending is permanent. In a workspace, these open from the New people row. The old name for all of this was Scout.

Try asking

• Find heads of data at Series B SaaS companies and email them about our tool
• Find me a VP of Sales who has sold software to banks, for my startup
• Not people from the big banks
• What happens when I sign off on a campaign?
• Run a campaign to 200 climate-tech investors
• Who replied to my outreach this week?
• Did anyone open the emails on this campaign?
• Turn on open tracking
• Why did my campaign stop sending?
• What changed on this campaign this week?
• Who on this list do I already know?
• Ask Priya for an intro to the 5 she knows
• Email the ones I know directly
• Skip everyone from my network on this campaign
• Add spintax to this email
• Vary the wording so every email is a bit different
• Spin the greeting and the closing on step 0
• Use this new copy for everyone we haven't emailed yet
• Add a second sequence with new copy for the investors
• Send the investors a 4-email sequence and everyone else 3
• Put the rest of the list on LinkedIn outreach
• How is each sequence doing?
• Move these 5 people to the investor sequence
• What happens if someone refers me to a colleague?
• Were these emails verified?
• Check these emails for deliverability
• Run a bounce check on this list
• What happened with the bounces?
• Find another address for the people who bounced
• A prospect's auto-reply gave their new address — did my first email go there?

When a search comes up short

How to

1. Nothing to switch on or choose: every people search does this when it needs more matches, and stops once it reaches your target. The old Ultrafind switch and Deep Search are gone; campaigns that used them search like every other now.
2. Your target doesn't change. People found beyond the index arrive in batches, usually within a few minutes, and meanwhile the search shows its progress on its own line.
3. See what it found beyond the index: when a search finishes I say how many our index found and how many more turned up beyond it, the results card shows the same split, and in the table each one found that way is marked. Ask me “how many did it find beyond the index on @campaign?” for the numbers.
4. Want it to dig deeper still? Tell me what to loosen or add to the criteria. When a search ends short, any suggestion I make is about the criteria.
5. Not the same as Scout's “New people → Both sources” refill option: that one only changes how Scout refills a campaign later.

Try asking

• Do I need to turn anything on to find more matches?
• Why did it find more people than the index?
• Can the search try harder?
• My search came up short — can you find more people?
• What happened to Deep Search?

Sequence templates

How to

1. Say “save these emails as a template” on a campaign, or “draft me a 3-step follow-up sequence I can reuse”.
2. Load one by its name, like “use my investor template on this campaign” (a list works the same way); it loads on your ask, with Undo, and you see what the campaign now sends.
3. “Add wordings” on a saved template wraps short phrases in {wording one|wording two} so each recipient gets one of them — your own wording stays the first option, and you can edit them all afterwards. Same thing as asking me to “add spintax”.
4. Ask “how are my sequences doing?” for per-template contacted/replied (plus opens and clicks when tracking is on in Settings → Campaigns).
5. Settings → Campaigns → Tracking is also where open/click tracking is switched on for ALL campaign sends, not just template ones — it applies from the moment you flip it, and earlier sends can never be backfilled.

Try asking

• Save this campaign's emails as a template called VC follow-up
• Use my investor template on this list
• How are my sequence templates performing?

Saved enrichments & criteria

How to

1. Say “always research check size on my searches” or “never include recruiters in my searches”, or open any campaign or /network column's menu and pick Save as enrichment.
2. Each saved item has a mode: Always (an enrichment added to every new people search), When relevant (I add it when a search fits — the default), or Manual (one tap on the plan). A criterion is never Always: it changes who is found.
3. After a search, add your saved enrichments to any campaign or list from the Research button, or ask me “add my usual enrichments to this list”. On /network they're in the Add column menu.
4. Manage them in Settings → Campaigns.

Try asking

• Always add check size and recent investments to my searches
• Never include recruiters or staffing agencies in my searches
• Add my saved enrichments to this list
• What are my saved enrichments?

Opportunities & campaigns

How to

1. Describe the goal; I draft an inline plan card you can refine and approve.
2. Approving spawns the campaigns/primitives; then ask me to check status, add or remove people, or launch.

Try asking

• Draft a plan to source and hire a founding engineer
• What's the status across all my campaigns?
• Add these five people to the campaign

Messages I send you and when

How to

1. HOW I DECIDE: for each message I would send on my own I ask whether a smart assistant would interrupt you for it — is it time-sensitive, does it need you soon, does it matter to you — weighing what you have already had today, how you have responded to that kind before, and how often you told me to reach out on my own (When it matters or Keep me posted; Only when I ask means I never start one). If no messaging app of yours gets the daily summary, nothing is held for it: my own messages go at their old pace instead. If it can wait, it waits for the summary.
2. NOTHING IS DROPPED: what waits comes in your daily summary each morning ("Also, from earlier", one line each, the first three then how many more); an update on someone I watch comes in your watch digest, and a reply waiting on you in the digest's "ready for you" line.
3. WHAT IS NEVER HELD: replies to what you asked me, results of tasks you started, your own watches and alerts, the recap of a call you took, word that a bug you reported is fixed, and urgent mail (a fraud alert, a meeting that moved) — those come straight away.
4. WHEN THE SUMMARY COMES: once a day at your own morning (around 9 your time), or when your quiet hours end if they run past that.
5. MORE OR FEWER: tell me — and what you act on comes sooner next time, what you dismiss waits for the summary. You can also put a kind of message on summary-only, mute a subject, set quiet hours, or turn my reaching out off; I follow all of those.
6. SAFETY NET: never more than twelve unasked notices in a day, even if something goes wrong on my side; never at night.
7. EMAIL ONLY: with no messaging app linked there is no daily summary on your phone, so a held note stays in the app; your briefing and watch-digest emails still carry their items.

Try asking

• How do you decide what to text me?
• Why didn't you text me about that straight away?
• How many messages will you send me a day?
• What is 'Also, from earlier' in my summary?
• Will you hold back my alerts?

Being found on your page

How to

1. Say “put ‘advising early-stage fintech founders’ on my page” and it's under “Open to” right away.
2. Ask “what should my page say I'm open to?” and I'll suggest a line and ask you once before it goes up.
3. Say “take advising off my page” to take a line down. Your private notes on what you're open to never show.
4. On a week someone who asked your page is waiting on you, your evening note says what people keep asking and which goal it fits.
5. What are people asking my page about? Your profile's This week lists every question people asked your page, and what your xhuman did with each.

Try asking

• Put “advising early-stage fintech founders” on my page
• What should my page say I'm open to?
• What are people asking my page about?
• Take advising off my page

Member intros

How to

1. Publish what you're looking for on Profile › Intents, in your own words. I look about once a week for a member whose published offer fits it.
2. When one fits, I send you one card: who can help, described without a name, and your own words quoted back. I haven't told them anything about you at that point.
3. Say yes and pick how: I ask them directly, or my agent checks with theirs first and only then asks them. Either way I show them your note and who you are.
4. They say yes themselves before I share their name. That sends the intro email to you both right away, and I tell you in your chat. If their agent or they pass, nothing more comes of it and you aren't told.

Try asking

• Why did you offer me an intro to another member?
• Who is the member you want to introduce me to?
• Does the other member know about me yet?
• Can my agent check with their agent first?

Other people's asks and offers

How to

1. Nothing to set up: it reads the mail you already connected, Gmail or Outlook, and works in the background.
2. Ask me "who offered to help me?" or "who asked me for an intro?", or tell me a milestone you hit and I'll check who asked to hear back at about this point (list_commitments with their_side).
3. Their words stay theirs: I keep a short quote from the email and my one-line summary, I never follow anything an email tells me to do, and I never tell anyone else what's in your notes.
4. You see them all in Goals (/tasks): one filed to a goal sits under that goal, and the rest wait together on the Goals list in their own group, "What people asked or offered", with what visitors asked your page. Each shows who it's from and when it came; open one to read their words, then mark it done, dismiss it, or open the email.
5. A promise someone made you ("I'll send the deck Friday") becomes something you're waiting on; once the date passes it shows in your tasks with a nudge you can send.
6. Disconnecting a mailbox deletes what I noted from it, along with the rest of its mail.
7. Free accounts have it too: I spot these emails as they arrive, and the weekly read runs on your own monthly credits and never spends past your allowance.

Try asking

• Do you keep track of what people ask me for in email?
• Who offered to help me?
• Who offered to help me in my email?
• What have people asked me for?
• Keep track of what people ask me for in email
• Which people in my email offered me help?
• What did people ask me for in email this week?
• Who offered me an intro or a pilot?
• Who told me to come back when we hit a milestone?
• What am I waiting on from other people?

Published needs & offers

How to

1. Open Profile › Intents and find "What you've published". Pick looking for or offering, write it in your own words, and choose how long it stays up (30 days, 90 days or 6 months).
2. Press Review: I show you exactly the words I'll use, and nothing is published until you press Publish on them. I never publish one for you, and I never reword it.
3. I read it when I look for who could help someone who asks the network. A published offer can make you someone I consider for any kind of ask; a need tells me what you're after when I do. On a free account, an ask that fits comes to you by text: YES shares your name with that one person and NO tells them nothing. On a paid plan your words count when I look, but those asks don't come to you by text yet.
4. Withdraw one on the same page and I stop matching on it at once. One ends by itself on its date, and up to 10 can be live at a time.

Try asking

• How do I tell the network what I'm looking for?
• Can I publish what I can offer?
• Who sees what I published?
• How do I withdraw my offer?

Booking the meetings you reply to

How to

1. Nothing to set up beyond a booking link: the reply alert says when it will copy my scheduling address.
2. Say "no cc" (or take my address off the card) and the reply goes without it; their pick then comes back to your inbox and I'll ask you before booking it.

Try asking

• Will you book the time they pick?
• Why is your scheduling address copied on my reply?

Communities and cohort codes

How to

1. Go to Settings › General › Who can see you › Communities you joined, type the code you were given, and press Join.
2. To leave a cohort, press Leave next to it there. A company follows your team seat or your work email, so it isn't left from here.
3. Each one says how you joined (a code, your team seat or your work email) and shows its size as a range ("10 to 24 people"), never an exact count; one smaller than five shows "fewer than 5 people".
4. Your own private lists of people (say, the VCs you're talking to) aren't communities: they're circles in Network › Trusted. Only you see a circle, and nobody in one is told.

Try asking

• How do I join my cohort with the code?
• Which communities am I in?
• How do I leave a cohort?

Follow-ups on quiet deals

How to

1. WHAT I WATCH: conversations with people where you both wrote — I read the thread, who they are to you, your goals and anything that happened after (another thread with them, a call and its notes).
2. WHEN IT'S DUE: I decide from the conversation itself — a warm reply you owe, or a nudge when you answered and they went quiet — and I set my own time to look again when it's too early.
3. ON YOUR HOME: one card per person, with the follow-up drafted as a reply in your thread, from the mailbox they wrote to. Send is one tap; Not now puts it off; Never stops cards about that person.
4. YOUR GOALS: each follow-up names the goal it serves, so calls and replies tie back to what you're working on.

Try asking

• Which deals went quiet that I should follow up on?
• Did anyone reply positively that I never followed up with?
• Remind me to follow up after calls

Goal research

How to

1. How often: about once a week for each goal, and again when you add a goal or move one up or down. I research your top goals, ranked the way you ranked them: three on a paid plan, one on the free plan or a trial.
2. For each goal I read who in your network fits that goal, what they are to it and what they might care about, with the line from your mail or notes that shows it. I also look for companies that fit, from the ones you know people at and from a search on the web.
3. When someone is worth reaching out to now, I put a card on Home with the email written to them, and the card says why I suggest reaching out to that person for that goal. For a company where you know someone, the card asks that person to introduce you to the right person there.
4. For a company that fits where you don't know anyone yet, I put one card on Home under that goal, with why it fits. Tap Find the right person and I make a list of who to talk to there, on your credits, and nothing is sent; or tap Start a search and I look for more companies like it.
5. If someone already has an open card from me, I don't write a second one. If you said no to someone or to cards like these, I leave them out.
6. When new research lands, your daily brief carries one line with the names and the angle.
7. It runs in the background on your credits, and each look stops at a small cap.

Try asking

• Who should I reach out to for my goal?
• Which companies should I partner with?
• Why did you suggest I reach out to this person?
• How often do you research my goals?
• Who in my network should I reach out to for this goal?

Small groups

How to

1. Publish what you're working toward on Profile › Intents, in your own words. When I review your goals each week, I look for a few members whose published words point the same way.
2. If enough of them do, I send you one card with your own words quoted back. I don't name or count them, and I haven't told them anything about you.
3. Say yes and I ask each of them privately, in their own chat. They see only their own words and that a small group is forming, with no names.
4. If at least four of you say yes within a week, I introduce everyone, book an hour you're all free and put it on your calendars. If not, it quietly ends and nobody is told who was asked or who said no.

Try asking

• Can you start a group for people working on the same thing?
• Who else is in the small group?
• What happens if not enough people say yes to the group?
• Do the others know I was asked to join a group?
• Can you set up a learning circle with other members?

Later steps

How to

1. Ask for the whole chain in one message: who each later email or invite goes to, and what it's for.
2. Read the card's last sentence: that is exactly what your yes covers.
3. To stop the later steps, tap Take it back on the task in Tasks.

Try asking

• Will you do the later steps too?
• Once she picks a time, can you email Leo the deck as the next step?

Find me opportunities

How to

1. Tell me what you're looking for; I ask a couple of gap questions, then scan.
2. I surface matches you can act on or track.

Try asking

• Find me opportunities in AI infrastructure
• What openings fit my background?

Job match

How to

1. Ask me to find roles that fit, or start a job-match feed.
2. Review matches; I can score and shortlist them.

Try asking

• Find remote staff engineer roles that fit me
• Start a job feed for product design jobs

Hiring

How to

1. Tell me the job and what you know: full-time, part-time or a contract, the hours, the pay, where. I write the whole posting in my reply; anything you didn't say, I fill with a sensible default and tell you, so you can change it.
2. Want changes to the job post? Just say so and I rewrite it. Tell me where it should go and I'll post it on that site in the browser, or post it yourself.
3. To find people for the job, ask me to find candidates: I show you who fits, and reach out to them when you say so.
4. xhuman has no job board or applicant pipeline of its own.

Try asking

• Draft a part-time bookkeeper posting
• Post a job for a part-time designer
• Help me hire a bookkeeper
• Write a job post for a full-time engineer
• Find candidates for a senior backend engineer job and reach out

Research studies

How to

1. Tell me what you want to learn; I walk you through goals, audience, screener, and questions with intake cards.
2. Launch it to distribute and gather responses.

Try asking

• Set up a study on how PMs use AI tools
• Run a quick survey with 50 designers

Ask people questions: surveys, pulse checks, studies

How to

1. Name the people by their xhuman handle and say what you want to learn; I write the questions and put the plan on one page — no setup steps.
2. Say yes and it goes out: each person is asked once, in the chat they use most, whether they want to take part, then one question at a time, with a follow-up when an answer is short.
3. Their answers land on your page as they come in. Make it anonymous and the page shows only how many answered.
4. Ask me how it's going, or say “stop asking” to close it.
5. People without xhuman, and paying people for their time, aren't available yet.

Try asking

• Ask @ana and @ben 3 questions about onboarding
• How many people answered my survey?
• Run a pulse survey with my team

Performance reviews

How to

1. Ask to run reviews; I set up the cycle with step cards right in chat — roster, org chart, review types, schedule.
2. Launch from the preview card; invites and reminders go out on schedule.
3. For a lightweight recurring team check-in, ask for a pulse survey instead.

Try asking

• Run 360 reviews for my team
• Set up a weekly pulse survey for engineering
• What reviews do I still owe?
• Draft my self-review from my last 6 months

Expert calls

How to

1. Tell me what expertise you're after; I set up the request with criteria and screening.
2. Approve to source experts; I schedule the call and handle payment terms.

Try asking

• I need to talk to someone who scaled payments infra at a fintech
• Book an expert call on FDA approval timelines

Interviews

How to

1. Ask to start an interview; answer by chat or voice and it adapts as you go.
2. Ask about what the interviews have surfaced.

Try asking

• Interview me about my background
• Run a mock interview for a PM role

Your network

How to

1. Ask a plain-English question about your network (“who do I know at Stripe?”).
2. Ask me to research a specific person, find their email, or email a set of contacts.
3. For the full picture on one person, ask me to “tell you everything” about them or to “dig into” them — I pull their career, their public web presence and recent news, and if they’re a contact, your whole history with them. That one takes 1–3 minutes and runs in the background.
4. Ask “any updates on Ana?” or “what’s new with Ana?” and I check everything at once — your emails and meetings with them, anything the monitors have picked up if you’re watching them, the news, and a live web search if you aren’t. I tell you which of those I actually checked, so “nothing new” means I looked.
5. Add LinkedIn connections by asking me — I show the connect card (live sync on a paid plan, or the free Connections.csv export upload; the dialog shows where to download the file). Once they're in, I check them against your live searches that include your network and hold anyone who matches for your say-so. Re-uploading later refreshes companies and job titles.

Try asking

• Who in my network works in climate?
• Any updates on Pia Calloway lately?
• What's new with Jane Park?
• What's next with Honor Poulton?
• Tell me everything about Jane Park at Acme
• Brief me before my call with Marc tomorrow
• Find the email for the CFO at Acme
• Email everyone I know who's a founder
• How do I import my LinkedIn connections?

Company research

How to

1. Ask in your own words (“which companies would make sense to partner with to sell to wealth managers?”) and I come back with up to ten companies, best fit first — or up to twenty if you ask for more.
2. Each company comes with how it fits what you asked and a reason quoted from its own site or profile. I say what that source shows and nothing more, so a company I can't back up with a quote isn't on the list.
3. For each one I name your contact there — the person from your own contacts with the warmest tie — or say you have nobody there yet.
4. Say if you want just companies where you have contacts, or just new ones from the web. The web searches carry the words of your ask and plain words like “partners” — nothing private about you or your contacts.

Try asking

• Which companies would make sense to partner with for selling to RIAs?
• Who sells to RIAs that I know someone at?
• Find companies that could resell our product to finance teams

Swap networks

How to

1. Sharing your network is a SWAP, both or none: say “swap networks with Ana”. Their own agent asks them once whether they'd like to swap with you, so you can each search who the other knows. A yes turns it on for both of you and you hear it in one line; a no is never told to you, and an ask nobody answers lapses after 14 days. If you two aren't connected yet, I connect you first and the swap is asked after.
2. When someone asks to swap with you, just answer: “yes” turns it on for both of you, “no” declines and they're never told.
3. Stop anytime, from either side: “stop sharing networks with Ana”. It's off for both of you at once, they aren't told, and their next look at your network just reads “no longer shared”. Disconnecting or blocking ends it too. When a swap ends, what it gave you ends too: people from their network you never acted on leave your campaigns, anyone held because of their network goes to your own review and stays held until you choose, and the “known by” note and warm routes through them are gone. Nothing is sent because a swap ended, and if you swap again they come back as they were the next time that campaign looks through their network.
4. Choose WHAT is searchable under Network sharing, the last row on Network › Trusted (/network?tab=trusted#team): LinkedIn connections, email relationships from all or selected mailboxes, and optionally Browse so they can scroll your whole people table (emails stay hidden there). Each trusted person's sheet on the same tab shows where your swap stands.
5. Everyone at your company is a pool: say “join my company's network pool” (or switch it on under Network sharing, on Network › Trusted). You and every colleague on your company domain who joined can look through each other's networks; a colleague who hasn't joined can't see yours, and you can't see theirs. Personal email domains have no pool. Leaving ends it both ways.
6. Once you and a teammate (or anyone you trust) have swapped, every network question I answer sweeps your network first and then theirs automatically. Say “search Alice's network” to search only hers, or open /network?owner=… to browse it. Any account can swap, free ones included, and looking through a shared network spends your own credits, like anything else you ask me. Their imported LinkedIn connections come in once your plan covers imported connections, the same as yours do.
7. You can bring teammates in while creating a campaign: tick “My team's network” on the “Where should xhuman look?” step and invite them on the teammates page. Each Invite click sends that one email right then — from your name followed by “via xhuman” (replies come to you), never from your own inbox — and the letter names the campaign and asks the teammate to swap networks with you for it. A teammate who joins through that invite gets a free account with 1,000 credits a month; nothing is searchable either way until they say yes, and until then the campaign says “no swap yet”. Nothing about invites leaves at Launch.

Try asking

• Swap networks with dana@northwind.com
• Stop sharing networks with Marc
• Can my colleague still search my network?
• Why can't I search a teammate's network?
• Does everyone at my company see my network?
• How do I let a colleague search my network?
• Search Alice's network for anyone at Stripe
• Who has swapped networks with me?
• Priya invited me into a team campaign — what does she see?

Next step with everyone

How to

1. Open anyone — from your people table, a chat card, a campaign, anywhere: under their name it says the one scheduled step — “Next: check in · Nov 21” — and its “…” changes the date or removes it. With nothing scheduled, “Set a next step” is in their “⋯” menu. Type the date in words — “in 3 weeks”, “next monday”, “nov 7”.
2. Promise a date in an email — “I'll check back in a couple of months” — and the send records it: the person's record shows the check-back date and I remind you when it comes.
3. When a contact has gone quiet with nothing planned, I look for a real reason to write to them now — news about them, something they asked for or were promised, a goal you share, a moment in their life or work. Only when I find one does a check-in show up in your suggestions, with that reason and the email already written; approving it sends exactly that email, from the mailbox you last used with them (in that thread when the conversation is recent). Tell me “never” about someone and I won't suggest them again. No reason, no check-in — I don't suggest a generic “just checking in”, and a check-in never gets a made-up due date.
4. On a morning a step you set or a promise you made is due, I read your conversations, meetings and notes with each of them and draft the email for the ones worth writing to — built on what was last said, never a generic “just checking in” — and send you one short message saying who I drafted for and why. Nothing goes out until you approve: the message shows each draft whole, and a 👍 on it (or Send on Telegram and WhatsApp, or SEND by text) sends exactly what it shows — or open its card in the app to edit first. I skip anyone you've been in touch with since, and if nobody's worth a note you hear nothing. Any email or meeting with them settles it.
5. That morning read is an agent turn on your credits (one per morning that has anything due); if you're out of credits you get the plain list of what's due instead.
6. Ask me what's next with someone, or “who do I owe a check-in this week”, and I answer from the same dates the record shows — I never invent one.
7. It is on for every subscription. If it is ever switched off for your account, the chip and the morning drafts stop — promises already recorded, steps you set and revisit dates stay where they are.

Try asking

• What's next with Honor Poulton?
• Set a next step with Marc for early November
• Who do I owe a check-in this week?
• Why did you suggest a check-in with Tom?
• Why did you draft a note to Surat this morning?
• How do I send the check-in you drafted?

Find a person by email

How to

1. Give me the address — “what's new with dana@newco.com” — and I bind that exact person, not a same-named one.
2. If the company you mention differs from what I have for them, I tell you which one I found rather than stopping to ask.
3. When I offer you a few people and you answer with one of their addresses, I go ahead with that person.

Try asking

• What's new with dana@newco.com
• Catch me up on the person who sent this email
• Look up sam@example.com

Warm routes

How to

1. It runs by itself: once your inbox or LinkedIn is connected (and whenever a teammate shares their network), I check everyone on your lists and put a line under each name saying how you can reach them. New people are checked as they are added.
2. “You know them” means they have written to you, you have met, or you are connected on LinkedIn — and the first cold email to them is HELD until you choose: email them yourself, or skip. The same for someone in a teammate's shared network: ask the teammate for an intro, email directly, or skip. Accepting one of your own campaigns' invites doesn't count: that reads “Accepted your invite” and holds nothing.
3. A search maps these as part of the search, ahead of its sign-off card: each person found gets their routes in words, rated Strong (they replied, someone put you in touch, you've met), Good (a contact who worked closely with them for 2+ years) or Possible (the same school or employer in the same years). A model reads each shared employer or school person by person and drops one that means nothing. Only Strong and Good routes get an ask to the contact.
4. Someone NOT on a list: “how do I get introduced to Marc Lee?” starts with your own direct tie. A first-degree connection is the first way in — I say so first, dated, with whether you have any email or meeting on file, worded as a connection and never as a friendship — before anyone who might introduce you.
5. A second-degree route — “via Dana · Dana emailed them 3mo ago” — means one of your contacts and that person have emailed each other on a thread you were on. It is a way in. Ask me to “ask Dana for an intro” and I draft ONE email to Dana covering every person on the list she might help with, for you to approve — nothing is sent until you click Send.
6. An intro ask (to Dana, or to someone who worked or studied with them) is ONE email from your inbox that asks whether they've met each person — never claims it — and ends with a short note to forward. Their outreach then waits for the campaign's intro wait (5 days unless you change it): an intro stops it for good, while a no, or no intro in time, starts it on schedule — and I say so in the campaign chat. Any other reply shows on those rows, which stay held until you decide.
7. With no list: “intro me to a person at their company”, or “yes, intro me” on a person I suggested — I draft ONE email asking your contact for that one intro, for you to approve; on your phone I print the whole draft and close on a question naming them. With no path to them I say so and offer a direct note. If they have not written back in a week, it shows on your Tasks.
8. In the background I fetch work histories for your 500 closest correspondents — mail both ways, or a meeting — so “worked with them” routes run through real relationships. Past 500, the campaign's people view offers more lookups. Two records of one person get joined.
9. “Jo worked with them at Fernhill Labs · 2019–21” means one of your contacts was at a company the person worked at, at the same time — read from both dated work histories, counting only the months the dates guarantee (at least six; a year at a company of over a thousand; fewer at a small company). It needs your contact's work history on file, from your LinkedIn connections' profiles; a contact you've only cold-emailed never counts. It never holds anything, and a shared employer is not a tie: ask me to “ask Jo about her” and I draft ONE email to Jo naming the company and the years, for you to approve.
10. “Michael was at Princeton with them · 2004–08”: a contact attended the same school in the same years — at least a year both education histories guarantee, so a short course never counts. A school many of your contacts attended is weighted down and called a loose tie. It never holds anything; ask me to “ask Michael about him” and I draft ONE email naming the school and the years.
11. Routes through a connector you correspond with — mail both ways, or a meeting — beat the same evidence from one you've never emailed, or whose mail you never answered, and the route card says so.
12. Each person's row says their best route in a few words — “via Arlo · long shot”, “No route yet · cold email”. Group the campaign's people by connector (“Arlo can get you to 2 of these”), by company (how many at each, and the single best way in) or, when it sends several, by sequence.
13. “LinkedIn 2nd · 12 mutual connections” comes from LinkedIn itself when your LinkedIn is connected live (a few profile checks a day, within LinkedIn's limits). “Ann works at Acme · possible” only means someone you know is at their company — they may not know each other.
14. Open anyone on the list and the route card on top draws the best way in — You → Arlo → Kasimir — each link in plain words (“6 years together at Larkspur Labs”), one word for how strong it is (Strong, Possible or Long shot; hover it for the exact chance), one sentence of why, and Arlo in full: face, role and your tie to Arlo (tap for Arlo's page, on top). “Ask Arlo for an intro” opens Ask for intros with the one note to Arlo. Beside a verified address a Long shot never leads: emailing Kasimir does. Other routes fold under “N more routes”; “Hide this route” in its “…” stops me leading with one.
15. At the top of the campaign: “4 of 5 are one intro away, through Arlo, Juno and Wes”, with Ask for intros — one short note per connector, each person asked about through their best one, each note shown whole for you to read, change or leave out. One yes sends every one of them at once; the line beside the button says how many go now and that they can't be unsent. Discover's Reachable chip narrows the list to the people someone can get you to.
16. When your inbox or LinkedIn lands (or a teammate shares their network), I tell you once in the campaign chat and on Telegram/SMS how many people are now one step away and how many first emails I held — nothing is sent to them until you choose.

Try asking

• Who on this list can I reach warm?
• Who can introduce me to Marc Lee?
• Which of my contacts knows people on this campaign?
• Ask Dana for an intro to the two people she knows
• Did Glen reply to my intro ask?
• Why are my warm paths mostly through acquaintances?
• Why does it say 'via Dana' under this person?
• Who worked with her at her old company?
• Only show me the people I can reach through someone
• Ask for intros to everyone on this campaign

Warm first, cold fallback

How to

1. It happens at the launch: when you ask me to launch, your ask is the go-ahead, and I tell you who I asked, for example: “I'm asking Ines, Kofi and Mira for intros to 4 people first. If there's no intro within 5 days, I'll email them directly.” Everyone else gets the campaign's first email as usual. In the campaign page's Send dialog, “Email them directly instead” launches with no asks at all. Before you send, you can say it for one person — “email Hugo directly instead of asking Ines first” — and the sign-off card shows it at once.
2. On a campaign that's already running, Ask for intros — at the top of the campaign, or “Ask Ines for an intro” on a person's card — shows one short note per contact, whole, for you to read, change or leave out. One yes sends every note you kept, all at once, from your inbox. From a person's card, a contact reached only on LinkedIn gets one LinkedIn DM instead. The people they ask about wait for the answer the same way.
3. Who is asked: only your own contacts who can be emailed — someone who writes to the person, or who worked or studied with them and writes to you — and each person through their best contact, once. At the launch, never a LinkedIn-only connection (from a person's route card you can ask one by LinkedIn message), and never a warm route you hid. A teammate is never asked at the launch; Ask for intros includes a note to them that you send yourself. At most 10 contacts at a time, and one note per contact covering up to 10 people.
4. At the launch, the ask leaves from your inbox right after it (or when a scheduled campaign starts), only while the campaign is sending: one note per contact asking whether they know each person — never claiming they do — with a short paragraph they can forward.
5. Say Ines can introduce you to Hugo. If Ines introduces you, or Hugo writes to you, Hugo's cold email is cancelled for good and the conversation carries on in your inbox. If Ines says no on the ask's thread, you hide the route through Ines, or the wait runs out, Hugo's first email goes out on the campaign's normal schedule — right away, inside the send windows and the daily cap, never all at once. Before anyone goes cold I check your mailbox itself, Gmail or Outlook. A LinkedIn DM ask waits the same way; Ines's reply there is the answer.
6. If Ines answers anything else (“let me check”), Hugo waits for you to decide: email them directly, skip them, or mark the intro made.
7. Each person's row says where they stand: “Intro asked from Ines — first email held until Oct 26” while the ask is out, and “Introduced by Ines — no cold email” once they're introduced.
8. “Fell back to email” means nobody introduced them in time, so the person's first email went back on the campaign's normal schedule. The row says why: “Fell back to email — Ines passed”, “Fell back to email — you hid the route through Ines”, or “Fell back to email — no intro from Ines within 5 days”.
9. At any time you can email someone directly instead, or skip them. There's no setting yet to change a campaign's wait or switch the asks off for a whole campaign once it's launched; to stop one, email the person directly or skip them.

Try asking

• Why did you ask Ines before emailing Hugo?
• When will Hugo get the campaign's email?
• Email Hugo directly instead of waiting
• What does “Fell back to email” mean?

Warm-path odds

How to

1. Each way in through one of your contacts shows a chance, e.g. “~15% intro, possible · Jo worked with them at Fernhill Labs”. It is how likely Jo knows them times how likely you can ask Jo — the number is the chance of an INTRO, not of how well they know each other. The drawer spells out both halves and the strongest evidence in one sentence, and I quote that sentence.
2. What raises the chance they know each other: they emailed each other on a thread you were on (the only evidence I call knowing — everything else is “may know”, whatever the number), the months both were at one company and how big it was then (two years at a 40-person startup says far more than two years at a 100,000-person bank), the same small company at different times (low), and the same school in the same years. Several pieces add up; the same company twice counts once.
3. What raises “can ask”: how well YOU know the contact — someone you write back and forth with is a far better ask than a LinkedIn connection you've never emailed.
4. The numbers are judgment calls until real outcomes come in — no one has measured them yet. Anything that isn't direct evidence says “possible” next to it and never goes above a cap. Asks, answers, intros and hidden routes are recorded to check them later.
5. People you know directly and your teammates' contacts are not a percentage: they come first, and whether a first email waits for your decision depends only on that, never on a number.
6. When I tell you a chance, it is exactly the one on the table — I never work out my own, I keep “may know” as “may know”, and I keep a LinkedIn count's date and that it names nobody. I say your contact's name, never “he” or “she”: the data says nothing about gender.
7. Asked who on a campaign you can reach warm, I rank every person on it by their best way in (list_targets with sort “warmth”) — I never judge from one page, and I never say the people I didn't read have weaker ties.

Try asking

• Why do you think Jo knows Marc?
• What does ~15% intro mean?
• How likely is an intro through Dana?
• Who's my best way in to her?

Person monitors

How to

1. Tell me who to follow; I set up a monitor and surface findings in your feed.
2. Ask what someone has been up to lately.

Try asking

• Follow these three founders and tell me when they post
• What has my competitor's CEO been saying?

Network watches

How to

1. Say what to watch for in your own words; I compile it, tell you who fits today, and flag the next one who moves.
2. Ask what I am watching for you, or why I flagged someone.

Try asking

• Watch my network for repeat founders with a $50M exit who start another company
• What are you watching for me right now?

News

How to

1. Ask what's in the news, or brief me on a topic.
2. Save stories and follow verticals you care about.

Try asking

• What's happening in AI today?
• Brief me on the news

Standing rules

How to

1. State the rule the way you'd tell a person: “from now on, always capitalize first names in outreach”.
2. I save it, tell you whether it's system-enforced at send time or followed as an instruction, and it applies from then on.
3. Ask “what rules do I have?” to review them; “pause”, “change”, or “drop” any rule the same way. Or open Settings › How it works for you (/settings/how-it-works#habits): under Habits it keeps your rules are grouped by what they're about, and each one's ⋯ menu has Edit words, Pause (Turn back on brings it back) and Remove — a paused rule isn't followed but stays listed. A rule that asks you first is listed once, under Always asks you first. What I noticed from your edits is listed apart, with Keep and Remove.
4. “Draft-only” (for this conversation, or everywhere, until a date if you say one): every email, reply, LinkedIn message or launch comes back to you as a draft instead of leaving, with no button that mails it, on the web, by text and on calls. What you click yourself, or scheduled before, is untouched. Tell me to lift it.
5. “Stick to my work and sales inboxes”: my inbox reads stay in those mailboxes, and every answer names the ones I skipped. A mailbox you name still gets read.
6. “Always ask me first” or “never” about a website or an automation isn't a standing rule — it lives in Settings › How it works for you, under Websites or Automations; about a person or a company domain it's your inbox's reply rule, which everything acting on its own follows (inbox, campaigns, invites, preps, automations, outside apps). Other standing rules are instructions I follow, not machine-checked.
7. Give a rule an end (“until Friday”) and it stops on its own. A rule that ends — dropped, undone or past its end — isn't silently gone: Settings › How it works for you lists it with the rules that ended, under What you've allowed, and I can tell you when and why, for 30 days; then it's erased.
8. Each instruction lives in one place: what you tell me as a rule is kept as a rule, not also as a memory — and dropping or rewording a rule forgets any memory that said the same thing. I name each one, and Undo puts it back.

Try asking

• From now on, first names must always be capitalized in our emails
• Always keep outreach emails under 120 words
• What standing rules do I have?
• Drop the rule about Fridays
• Keep this conversation draft-only — don't send anything
• Only use my work and sales inboxes
• Is my sign-off rule still on?

One preference, every place it's kept

How to

1. Change or drop a preference the way you normally would: edit an automation's instructions, or ask me to forget a memory.
2. Read my reply: it says what else changed (other memories forgotten, standing rules ended).
3. On the web, tap Undo on that change's card (or in the automation editor) within 5 minutes and everything that changed comes back. By text, I can put the instructions back, but the memories it forgot come back only from that Undo.

Try asking

• Drop the funding thing from my news brief
• Forget that I'm fine with funding rounds over $100M
• Why did my memory disappear when I edited the brief?

Outreach defaults

How to

1. Say the default: “default all my new campaigns to 40 emails a day”.
2. When you change one campaign's setting, I'll offer once to make it the default for future campaigns, existing ones, or both.
3. Defaults apply at creation — existing campaigns change via the matching apply-to-all action or per-campaign edits.

Try asking

• Default my campaigns to 40 emails a day
• Always cc my cofounder on outreach going forward
• What are my outreach defaults?

Inbox placement (spam test)

How to

1. Ask me to test your deliverability (or a campaign's first email) and I run it right away — one real test email leaves your mailbox.
2. Results land in the card in a few minutes; ask me for the report any time (“how did the spam test go?”).
3. If a campaign's reply rate looks like a placement problem I'll say so on the campaign read and offer the test — I never run one unasked.
4. You can also run and browse inbox placement tests under Settings › Email & calendar › Sending › Spam test.

Try asking

• Are my emails landing in spam?
• Test my deliverability
• Run a spam test on my Gmail
• Why is my open rate so low?
• Nobody replies to this campaign — is it the inbox?
• Check my SPF, DKIM and DMARC
• Is my domain blacklisted?
• How did the spam test go?

Engine

How to

1. Open Engine from the sidebar on the home page (or /orb?engine=1) to see every campaign at once; the campaign chips narrow it to one.
2. Each column is a stage: Pool (found, not contactable yet — anything waiting on YOUR decision is pinned at the top), Researching (a lookup actually running now), Up next (queued and approved, lined up the way they will go, batched by day), Reaching out, Replied, Meetings, plus a Closed tray.
3. Click a card to open that person; click a campaign name to open the campaign.
4. Come back after a while and the board tells you what changed while you were away.

Try asking

• Where can I see what the agent is doing?
• Who is next in the reach-out?
• What did you do overnight?
• Show me everyone being researched right now
• Which people are in today's batch?
• What moved since yesterday?

Outreach modes

How to

1. Approve each email — the launch is your approval of everyone already on the list. Anyone I find later waits in the review queue until you approve their first email, a batch at a time; approving a batch is the send decision for that batch, and it goes out when due. Follow-ups ride the sequence you approved; replies wait for you. A campaign that tops itself up the older way contacts strong matches under the launch approval.
2. Approve replies — emails, LinkedIn requests and messages go out in your name, first touches and follow-ups, under the daily caps and to verified addresses only (an automatic mode locks that on). New people are contacted the same way, up to the campaign's monthly credit cap when it has one (3,000 on a new one; without one it keeps finding people uncapped until you set one in the campaign's Limits). On a fixed list nobody new is added. Replies and meetings wait for you — a pick of your times comes as a confirmation whose sending books it (LinkedIn: you're told).
3. Just run it — Approve replies (daily caps, verified-addresses-only) without showing you anyone first: it answers prospects in your voice, on its own, without showing you the reply first, and puts the meetings someone accepts straight into your calendar. I am the one writing all of that, and I can get it wrong; it goes to real people, under your name, and a sent message cannot be unsent. Choosing this mode is you taking that on. Approve replies does the same finding and contacting and still hands you every reply and every meeting.
4. How Just run it books — a meeting only counts once it is on both calendars. When they take ONE of your offered times, or name ONE free time of their own inside your working hours and at least four hours out, the calendar invite goes to them first, then my reply confirms the time and joining link — if the invite cannot go, nothing is booked and I tell you. I repeat the time in their time zone (a zone they named, like “3pm ET”, stays theirs). A maybe or several options comes to you as a draft; a pick with a question only you can answer, as a confirmation for you to finish.
5. Before anything I wrote goes out, or a time is booked, unread, a second check asks: did they agree to exactly that time, does it say anything you never said, is it in their first outreach's language? Anything short of a clear yes comes to you — as does a request for a deck, pricing, documents or facts only you have, unless your approved answers and files (Settings › Files) cover EVERYTHING they asked. Contracts, legal and security always come to you.
6. If they ask to add someone, like a cofounder, they join the invite after the same checks; an invite meant for someone else, like an assistant, comes to you. If they ask for your booking link, they get it (with none of that length, by email they get times and you are told; on LinkedIn it comes to you). A picked time that filled up or passed, or a proposed one that isn't free, gets fresh free times once (the closest, for a proposed one); a second one comes to you. If they decline every time but name a window, I offer times inside it, never declined ones; none free, it comes to you.
7. On LinkedIn it works the same way: I answer their messages and book a time they take — but first I ask for the best address for the calendar invite, or confirm the one on file, and hold the time; it counts as booked only once the invite has gone there, and I confirm in the chat where it went. If they never answer I ask once more, then tell you and free the time; if they would rather not share an address, or ask for your link, they get your booking link instead and nothing is held. I follow up once there too.
8. Once a meeting is booked, if they ask to cancel or move it, I do it myself — cancelling it, or moving that same meeting (with no new time named, I offer times first) — and tell them and you, once per meeting; anything unclear comes to you. A declined invite comes off your calendar without a word to them (an Outlook decline comes to you); a new time proposed on it is a move. A day after a cancel with nothing rebooked, I ask once if it is still worth meeting, unless they said no to another time. An opt-out leaves a booked meeting in place, and you are told. Below Just run it the reply waits for you (LinkedIn: you're told).
9. There is no limit on how many replies I send one person in a campaign. A loop stops me — the same long message from them again, one with no words of their own, one written by an automated system rather than the person, or eight of my messages to them within a day — and it comes to you with the reason. A reply 14+ days after the campaign's last message to them, with no meeting ahead, is yours: I tell you the campaign won't act on it (an opt-out still counts); by email I watch for those about two months. We can switch any of these actions off for a while (it then comes to you, and I say so); to stop them yourself, lower the dial to Approve replies.
10. The one follow-up: when times went out, or they said they would book one, and about two business days pass with nothing on your calendar, one short follow-up with fresh times goes in their working day, in your voice and their language — on its own on Just run it when I sent the times, otherwise drafted for you (on LinkedIn I tell you it is yours to send). Times I offer on my own start the next business day at the earliest. If no time fits their day, or I could not write it in your voice, it comes to you.
11. After the follow-up: about three more quiet business days and I tell you and stop — never a third message. A reply that only says they will pick a time later gets no answer and restarts the two days (after a follow-up, the next is yours); anything new — their answer, you writing to them, someone else in the thread, a meeting booked any other way — stops it.
12. Finding people is its own switch, “New people”: off, nobody new is found in any mode; turn it on and it starts adding people again.
13. Which language I write to prospects in: the language of the first thing THEY were sent (a segment with its own copy counts), kept for their conversation even if you later rewrite the first step. Ask me which it is.
14. On campaign threads one reply never waits, in any mode: when someone emails back asking to be taken off the list, the address is blocked and a single short line acknowledging it, in your voice and their language, goes back into that thread from your own mailbox, with no card and no draft — or nothing, if I can't write it in your voice and their language. Below Just run it, it is the only reply sent in your name without you. LinkedIn has no equivalent acknowledgement.
15. The dial: moving it up asks you to confirm first, moving it down takes effect at once, and Undo lasts 5 minutes after a change (undoing a move down asks you to confirm too; undoing a raise also stops the finding it started).
16. In chat: no card needed for a move down; I never undo it back up myself. Approve replies I set when you ask, and a campaign's Just run it too (in the app, or typed on iMessage, Telegram or WhatsApp), with no card; if it is already there and you ask me for autopilot, I leave it. Asking me to keep it topped up turns finding people back on and takes the campaign off Pause everything (I tell you how many and when they leave).
17. Old names: Scout → New people and Limits on Campaign rules; Review → Approve each email; Autopilot → either automatic mode (Approve replies or Just run it); Pause all → Pause everything.
18. Only the campaign's owner can raise the mode, turn an automatic mode on, or resume one from a hold, never from an automation; an editing collaborator can lower it.
19. Where I find the people, an automatic mode first checks the sending domain (missing SPF or DMARC, or records unreadable just then, keep the dial put and name the record to add), sends at its normal daily pace from day one, and pauses itself if two of its first 20 hard-bounce; turning it on again starts a fresh count.
20. Editing who it targets, the sender, or the emails themselves (a step, the email plan, a cc, an attachment) never pauses an automatic mode: it keeps sending on the new settings (after an email edit, the edited version), to people the launch approved too — nothing waits for your re-confirm. Where I find the people, a lapsed subscription puts an automatic mode ON HOLD: people found since the launch and not yet contacted move to the review queue; people the launch approved are still emailed, and follow-ups keep going.
21. A hold keeps the mode: the owner resumes it on the dial or the hold card in chat; no re-confirm is needed. Resuming does not empty the review queue.

Try asking

• Run this campaign on its own
• I want to approve every email before it goes out
• What will it do without asking me?
• Why hasn't this campaign sent anything?
• Put this on autopilot but let me approve the replies
• What mode is this campaign in?
• Will it reply to people for me?
• Stop asking me to approve every person
• Will it follow up if they don't book a time?
• What happens if a prospect cancels the meeting?
• Will it keep replying to an auto-responder forever?
• What if someone replies a month after the campaign ended?
• Is there a limit on how many replies it sends one person?
• Can I turn off just the meeting rescheduling?
• What if they say none of the times work?

Answers & files

How to

1. Add them in Settings › Files, or tell me in chat (“save this as my pricing answer: …”) — it's saved as you asked and replies use it from then on; ask me to change or remove it any time. I don't save one from a suggestion you tapped or on an automation's turn. A file is a PDF of up to 15 MB in Settings › Files with Campaigns can send turned on; you give it a plain name for messages (“our one-pager”).
2. When someone asks about pricing, asks whether it does X, or asks for a file you approved, I check whether your approved answers and files cover EVERYTHING they asked — all or nothing. If they do, on Just run it I answer from those alone and it goes on its own, by email or on LinkedIn. Below Just run it, by email the reply is drafted from them, with the file on it, for you to send; on LinkedIn, which has no drafts, you're told they wrote and you answer in the chat. By email, if anything isn't covered I tell you which topic, and it's yours; on LinkedIn that check runs only on Just run it.
3. Every price, percentage, date or link in your answer goes out exactly as you wrote it — I restate your words in their language, but the figures are copied, never retyped — and a condition you wrote (“quotes above 25 seats come from me”) keeps the cases it names coming to you. A second check reads the reply against your approved words before it goes.
4. Files: by email your file is attached (up to 3 MB; above that it goes as a link), and on LinkedIn it's always a short link. A link shows your current version — replace the PDF on the file in Settings › Files and links already sent show the new one — and it stops working if you delete the file. A file is sent only when they ask for a document, never alongside the answer to a plain question. Security documents (a DPA, a SOC 2 report) are never sent on their own, and contracts, legal and security questions always come to you — even if you saved an answer about them.
5. After a meeting is booked, a deck they ask for is still sent. When one goes out on its own you're told which answer or file it used, on every channel you've linked. Edit, replace or delete one any time: a reply written from the old version that hasn't gone yet waits for you instead. Changes you make in chat aren't part of undo — edit or delete them again, here or in Settings.

Try asking

• Can it answer pricing questions for me?
• Save this as my pricing answer
• Will it send my deck when someone asks for it?
• What are my approved answers?
• Add my one-pager so prospects can get it
• Why didn't it answer their pricing question?

Files

How to

1. Upload in Settings › Files (drag files in or press Upload): PDF, Word, Excel, PowerPoint, CSV, text and images, up to 25 MB each and 1 GB in all. Files you send me in chat show there too under Sent in chat; press Keep on one to keep it, or it goes 30 days after you sent it.
2. Ask in chat: “send them our deck”, “attach the NDA to the follow-up”, “what does our pricing one-pager say about seats?”. I find the file in your Files and tell you which one I attached.
3. Campaigns can send: turn it on for a PDF in its menu and say what people ask for when they want it and what a reply calls it. Campaign replies then send it when someone asks, attached by email (a link above 3 MB) and as a link on LinkedIn.
4. Each file's menu also has Download and Delete. Deleting can't be undone; for a file campaigns can send, links already sent stop working.

Try asking

• Where do I upload documents for you?
• Send them our deck
• Attach my NDA to an email
• What files do you have of mine?
• Read my one-pager and tell me what it says about pricing
• Keep the contract I sent you in chat

Who gets LinkedIn

How to

1. Decided at launch (or when you add LinkedIn to a campaign that is already sending) and re-balanced every week (and whenever new people are found): the account's weekly invites — ours, or your own setting — go first to people with no email (on every campaign of yours), then to the best fits by campaign priority (strong matches; on a list nobody scored, the people you approved, then as you added them). It applies to a campaign that opens with an email; one that opens with a connection request walks its steps as written. The campaign page says who gets LinkedIn now and who gets it later, and each person's record says which group they are in.
2. “LinkedIn later” people are invited once their emails went unanswered for 3 business days, best fit first, while there's room and the account can send. A reply, an opt-out, a bounce or a meeting ends it.
3. If the invites run out before a best fit's turn, they move to the later group and their next email goes on time. Their record says why; unskipping the step puts them back, even after their emails moved past it. Anyone whose invite already went is never moved.
4. A few messages a day are kept for replies and confirmations. If the network itself refuses invitations for its weekly limit, I hold the account to that number until its week rolls — that limit is LinkedIn's. When the week is full with people waiting, I tell you once a week on your linked channels.
5. Ask about any campaign — who gets what on this campaign, or why a person didn't get an invite yet — and I answer from that campaign's own counts and the person's own record.

Try asking

• Who gets LinkedIn on this campaign?
• Why didn't Ana get an invite?
• When do the rest get invited?
• Why are my invites waiting?

How long a campaign takes

How to

1. The launch says the finish in weeks — “About 7 weeks” — and, when one group finishes well before the other, which group that is. That sentence shows on every account; the four choices below it are the part that can be switched off.
2. Always the same four, in the same places: the DEFAULT, by fit — nothing to click (people with no email are invited first, your best fits get invites alongside their emails, the rest get email now and an invite later), Email only (drop the LinkedIn steps — anyone with no address is then not reached at all), Both for everyone — shown as “LinkedIn for everyone, by fit” (everyone invited at the account's pace, best fit first; an email still never waits for an invite; and the way back), Raise my LinkedIn limit (lift the weekly invite ceiling on the account and the campaign together). Each says who gets LinkedIn, who gets email, and how long.
3. On a campaign that opens with a connection request (or if the by-fit default is ever switched off), the first choice is “LinkedIn for the best-fit N” instead: it skips the invites for everyone outside the group, their record shows “Outside the LinkedIn group”, and you can put anyone back by unskipping the step by hand. That group is chosen ONCE, at launch — anyone found afterwards keeps their steps.
4. Changed your mind: choose “Both for everyone”. It puts back every LinkedIn step the group skipped — and never touches a step you skipped yourself.
5. Ask me on any campaign: “how long will this take?”, “make this go faster”, or “why is LinkedIn so slow on this one?”. I answer from the same arithmetic the card and the worker use, so the number you are told is the number that will happen.
6. It is on for every subscription. If it is ever switched off for your account, the weeks still show; you just are not offered the four choices, and no new group is picked. A group already applied stays applied until someone chooses “Both for everyone”.

Try asking

• How long will this campaign take?
• Why is this going to take seven weeks?
• Can you make this campaign go faster?
• Only invite the best fits on LinkedIn
• When will everyone on this list be contacted?

Sending capacity

How to

1. Ask what you're sending and I'll show the split per account: what's owed to follow-ups, what's left, and what each campaign is getting.
2. Set the ceiling for an account (“cap me at 50 a day”) — that's the pool everything is divided out of.
3. Rank the campaigns. The top one is filled first; a campaign below the line starts nobody new that day, though its follow-ups still go out.
4. Set one campaign's target with “10 first emails a day on my Q4 outreach campaign”.

Try asking

• How many emails am I sending a day?
• How many people can I send per day?
• What's my global max emails?
• Why isn't this campaign starting anyone new?
• Cap my sending at 50 a day
• Make the investor campaign top priority

Automations

How to

1. Describe the rule (“every weekday at 8am, summarize new investor replies”).
2. I turn it into a scheduled automation you can edit or turn off.
3. Say where the result should land — Telegram, text message (SMS), or email. Texts need the SMS channel switched on first.
4. A rule set to act without asking (autoSend) sends on its own through its own email and calendar steps: an email rule with autoSend answers the emails it matches in your name, and the rules that let me answer your inbox on my own stand down for those. A rule I run for you as the assistant may act on its own too: it sends what your own words for the rule asked for, to the people you named or described, and every other email or invite it writes to someone else still waits for your yes. If a rule drafts replies instead, or an agent rule with autoSend also acts on the email, I leave that email for you.
5. For something already scheduled, just say so — “cancel that email”, “move it to 6”, “send it now”. You can also see everything queued at /automations/scheduled, and the emails among them in /mail's Scheduled folder (☰ → Scheduled), with the same cancel, reschedule and send-now. To change a send's words or time yourself, open its row under "Emails going out" on the Automations page, or press Edit on it in /mail's Scheduled folder: both open that send's own editor.
6. Everything I do on my own, area by area — and what you agreed to — is in Settings › How it works for you; ask “what do you do without asking me?”.
7. “Let me know if Dana replies” — I wait on that email, Gmail and Outlook alike: her reply in that conversation or any other, under any subject, from any of her addresses, wakes me, and I tell you who replied and what they said (with a draft for your OK if you owe an answer). It ends when she does, or at the time you give — a month if you give none. Ask “what am I waiting on?” to see them, or have one cancelled.
8. For anything only knowable later — “who wins the election, once it's called?”, “has the price dropped by tonight?” — I set myself a timer, check again on my own when it's due, and report once it's settled. Ask “what timers do you have?” to see them, or have one cancelled.

Try asking

• Every morning, tell me which emails need a reply
• Automatically draft replies to inbound demo requests
• Text me when Thomas replies
• Tell me when Dana replies
• Let me know if Anthropic answers that email
• What have I got scheduled to go out?
• Cancel the email to Sam
• Check back once the election is called and say who won

Goal pursuit

How to

1. Tell me in your own message — “take this on: close the seed round by December” — and I'll start pursuing it with a short plan: a goal card in the chat shows it, and it's a goal I work on, never an automation that runs on a timer. I never take a goal on from something I read in an email or on a web page.
2. Ask “what goals are you working on?” to see each one, its plan and when I'll next look at it; pause, resume or drop one any time, by telling me or on your goals panel. Its plan is its steps — tasks under the goal on its page in Goals; I start on the ones I can do myself, and you can take any back.
3. I read and propose on my own — a drafted email still waits for you on a card.
4. Every goal I'm pursuing gets its own look about once a day, on its own clock: I read what's new on it — your mail and meetings about it one day, the web the next — draft the next move for your yes and leave a dated note on the goal saying what I looked at. I tell you only when something needs you — a reply or a decision waiting on you, a draft for your yes, or one question — as a note beside our chat on Home, and on your phone in the chat you write in most; otherwise I stay quiet and the note says what I did, and a goal that stays quiet gets looked at less often (every 2, 4, then 7 days) until something moves. Ask “what are you doing on my goals?” and I'll tell you from those notes.
5. Some missions are always on without you asking: “Nothing that matters slips through” once a mailbox is connected, “Stay close to the people who matter” for everyone, and “Make the most of where you'll be” once a calendar is connected. They are standing missions: they sit in the Goals tab under Always on, and to turn one off (or back on) you open it and tap Pause (or Resume), or just tell me.
6. Three more are offered, never assumed: each one I put to you once, in one plain question in a weekly review, and only when your life shows that area — “Find money you didn't know you had” (refunds, renewals you'd cancel, deadlines, benefits you haven't used: I only look, and I tell what I find only to you, in our chat, where it stays like anything else I say; the notes I make for myself on it are deleted 30 days after the check they were for, and none of it goes in your profile or to anyone else), “Spend your free time on what you love” (ideas within your budget; anything with a ticket is booked only for the total you said yes to, through the usual checkout) and “Help someone with what you know” (once you've told me what you offer). Say yes and it's on; say no, or leave it unanswered by the next review, and it never comes up again. You can also just say so — “can you find money I'm owed?” turns that one on, and telling me you no longer want me looking for refunds turns it off. “What are you looking for on your own?” gets every mission I keep looking after for you.
7. When nothing has moved on a goal for 5 days, my next look doesn't stay quiet: I suggest one new path to get it moving, with what it costs in credits, or ask you one question. The goals list reads “Stalled 5 days · suggests a new path”, the suggestion waits in Needs review as one decision, and on your phone it comes as one message: say go and I start it as a step on the goal.
8. When something happens that matters to one of your goals — a reply on its thread, new mail about it, a check I set for it coming due — I look into it within minutes, for that goal only: I read what came in, draft the next move for your yes and set my next look. I never send, pay, post or delete on my own for it, and if it doesn't move the goal I stay quiet and just note on the goal that I looked. New mail is weighed against the goals you told me yourself, not the always-on ones.
9. Ask “how do you count the luck you made me?” and I'll tell you: every item I count is real (a lead I found, a reply, a meeting, an intro, a done task), and I sort each into three kinds — found (I spotted it and you hadn't acted: a watch hit, a warm intro, a lead), made (I did the deciding step on your yes: a scheduled send, a booking, an intro request) and sped up (a reply or nudge you'd likely have handled anyway, like an inbox reply ping). Found plus made is the luck I made; sped up counts as activity but never as luck. It's grouped by goal, and by area of your life when a goal has one. I only count what actually happened, and never cards nobody saw.
10. Steer what I work on just by telling me: “focus on the raise this month” makes it my top priority, “done when I have an offer in writing” tells me what finished looks like, “waiting on Dana until Friday” tells me to hold until then, and you can tell me which part of your life a goal belongs to. “What are you working on for me?” gets them by priority, with each goal's area and when I'll next look.
11. “What do you look at for my luck?” lists it: each mailbox, each calendar, your contacts, LinkedIn, call notes and company knowledge — all on by default. Turn any one off with its switch (each mailbox and calendar on its row in Settings › Email & calendar, LinkedIn on its row in Settings › Apps, and the rest under Also reads there), or just tell me (“stop looking at my work inbox for opportunities”), and I stop reading it when I look for opportunities for you.
12. Saying yes to a lot of my drafts doesn't switch on sending by itself: replies and first notes to new people still wait for your yes however many you've okayed as written, and I don't offer to stop asking about them. I do keep count of the ones you approve without changing a word. What sends without asking is only what you chose: a campaign's mode, or a rule you gave me for your inbox.
13. Ask “what are you trying for me?” and I'll tell you my bets: for each goal, the one to three ways I think it could happen — what I'm betting, why, how I'm testing it, what success looks like and my next step. A bet I suggested shows as my guess until you say yes or something it tried pays off. What works raises it, what you dismiss or mute lowers it, and a run of misses drops it; a bet you or the results killed stays dropped and I won't suggest it again. Tell me “stop trying that” to drop one, and you'll see each bet on its approach in Goals.
14. When a bet is about finding new people who fit — people you don't know yet — the weekly review can give it a people search, built on something uncommon you share with them when there is one (your school, a past employer, your city). You get one card for it asking whether to run it; your yes finds the people and sets up the outreach for you to review, filed under that goal and bet, and nothing is sent until you launch it.
15. When a people search for one of your goals finishes, you don't get a list of a hundred to review: I grade everyone it found with the campaign's own match grade, pick the best 25 I can reach now (an email on file, or a warm intro), draft the notes and ask one question with each pick's face and why. Your one yes reaches exactly those picks, Edit opens the notes, Not these skips the batch, and the full list is one tap away on the campaign's page.
16. A bet is tested by watches of any kind — a listing, an event, a recurring search — each filed under its goal and that bet. When one sees a change I look into it for that bet only, and you hear the substance or nothing. Dropping the bet stops its watches; pausing the goal pauses them.

Try asking

• What goals are you working on?
• Take this goal on and keep me posted
• Stop working on the fundraising goal
• What are you working on for me?
• Turn off a standing mission
• How do you count the luck you made me?
• Focus on the raise this month
• What do you look at for my luck?
• Will you stop asking me about replies after a while?
• What are you trying for me?
• What do you do when something happens that matters to my goals?
• What are you doing on my goals?
• Can you find money I'm owed?
• What are you looking for on your own?

Goals plan themselves

How to

1. Confirm a goal in Goals (“Yes, that's my goal”), or answer yes when I ask about one — by text, Telegram, email or on the app — and I start drafting its steps; the goal says “xhuman is drafting the steps” while I do. Accept a goal card or a “Still working on this?” check-in and I start within a few minutes. Tell me a goal in chat and I lay its steps out right there.
2. My steps start on their own once you confirm the goal in Goals, answer yes to my question about it, or accept its card. A step I started says so on its row, and Take it back stops me and makes it yours again.
3. Steps we laid out together in chat wait: the goal asks you to “Pick which steps xhuman runs”, so open it, untick any you don't want, and press “Give N to xhuman”. Or give one at a time from its row, or select several and use “Give to xhuman”.
4. Steps marked yours stay on your list as your to-dos. The goal shows how many are done, what I'm working on and what needs you.
5. Tell me in chat to add, change or drop a step, or to plan a goal again.
6. Replies you owe and people you're waiting on, I do the work instead of reminding you: I draft the reply in their thread, from the mailbox it came to, and when someone hasn't answered you I draft a short follow-up in that same thread, to them alone. The message says who it is, what they want and why it matters, with the whole draft; one yes sends it. Say “not now” (or “not now, Thursday”) and I'll bring it back then — no day named means the next weekday morning at 9, and you can name another. Once your reply goes, they answer, or a meeting gets booked, it comes off your list by itself. On the free plan this runs on your own monthly allowance; drafting a reply you owe from scratch is still a paid-plan step there.

Try asking

• What happens when I confirm a goal?
• How do I choose which steps you work on?
• Where are the steps for my goal?
• Do you start on my goal's steps by yourself?
• Can you draft the follow-up when someone hasn't answered me?

Your luck

How to

1. “Do you review my goals?” Yes: once a week (once a month on the free plan, inside its allowance) I review every goal like a chief of staff would — my bets, how each went, the ones you turned down and anything waiting on you. I keep, push or drop each bet, suggest at most one or two new ones you'd likely never have thought of, and set every goal's next look. I only drop my own guesses you never said yes to; a bet of yours I check with you first. When nothing is ready I stay quiet and it costs nothing; otherwise you get one message with the bet, the change and the next step. “What did you change this week?” gets the last review's changes on each goal. Taking a goal on (“take this on”) brings its first review within about a day.
2. Ask “where can I see what you're trying for me?” — on Tasks, under your goals, in “Your luck”: your goals by area of your life, my bets, what came of them (found, made happen, sped up) and what I look at. “What has xhuman done for me this month?” gets the same count. Once a month I also text you that story, and sometimes a short “here's what I think you're after — right?”; with nothing to tell I stay quiet. Health, family, relationships and money stay out of that text.

Try asking

• Do you review my goals?
• What did you change this week?
• What has xhuman done for me this month?
• Where can I see what you're trying for me?

Page watches

How to

1. Give me the link and the condition: “watch this page and tell me if the price drops below $400.”
2. It shows under Watching for you with Pause, Resume, Check now and Delete.
3. Checks run as often as every 5 minutes (hourly by default, every 30 minutes for prices) and only on public pages — not pages that need you logged in.
4. If a page keeps failing to load (three tries in a row, or a day with no good read), I tell you once and stop watching it.

Try asking

• Watch this page and tell me if the price drops
• Tell me when this page changes
• Let me know when they post a new item on this feed

Watches

How to

1. Tell me what to watch and what counts as a change: “tell me when this status page is green again”, “let me know if the price drops under $300”, “check every hour whether the new version is out”.
2. I write the check, run it once right away to make sure it works, then it runs on its own — as often as every minute, or as rarely as you like.
3. While nothing changes you hear nothing, and no AI is used. When it sees the change, I tell you here, and on your linked messaging apps.
4. A check runs in a fresh, throwaway sandbox with the internet but none of your accounts, passwords or files — it can only read public things, plus one thing of yours it is handed to read: how many credits you've used lately, in total and per rule, task or automation.
5. That is how I warn you when something burns credits: “warn me if any rule uses more than 50 credits a day” becomes a watch that reads your own credit use (the last hour, day or week) and speaks up only when a rule passes your number. It reads only your own use and never changes anything.
6. Ask “what are you watching?” to see them, or “stop watching that” to stop one. If a check keeps failing, I stop it and tell you once.
7. Each check uses a few seconds of computer time from your credits; telling you about a change is one short reply.
8. A watch can work for one of your goals: I file it under the goal and the bet it tests (“watch the internal job board for my move into product”), so ask “what are you watching for my raise?” and I'll list them. When that bet is dropped or the goal is done, its watches stop; while the goal is paused, they wait.
9. While I work a goal on my own I may start a watch for it too — on the web only, at most hourly and three at a time, and only on a paid plan. You hear from it only when a change really moves the goal; otherwise it stays quiet.

Try asking

• Tell me when this status page is green again
• Watch the price and tell me if it drops under $300
• Check every hour whether the new version is out
• What are you watching for me?
• What are you watching for my raise?
• Stop watching that

Your twin

How to

1. Open /profile and tell me what's changed in the line at the foot — I'll update what I know about you.
2. Tap Edit on any section of your profile (your line, Working on, Open to, About, Background, Knows about, links) and tell me the change; each section also has its own Public / Private switch.
3. Ask me who asked your twin what, and about new followers and requests to connect.
4. Ask me to train your twin from your notes, or set its voice and signature.
5. Ask what visitors have been asking your twin.
6. Open Settings › Memory to see what I know about you, or ask what I think you're working toward — I'll list my guesses as guesses, and you can confirm, correct or remove each one.
7. When a suggestion says what it assumes about you, just reply — "no, I closed that round" or "forget that" — and I'll drop or correct that guess.
8. When I file something you tell me on your life map under money, family, health or the people in your life — or as something you find hard, or a limit you live within — it stays at Only you and xhuman: never said to anyone, unless you share it yourself in What xhuman knows. I never guess in those areas; anything there comes from your own words.
9. Once I've done something useful for you on your phone, I ask a few short questions over the next messages, one at a time: what you'd love to happen, what feels hard right now, what you could help other people with, and a little about your life. Say you'd rather not and I won't ask that one again. What you say on the 5-minute intro call counts too. Your answers go into What xhuman knows, with the same privacy as above.
10. Your page is your handle's link, built for a phone: your photo, name, handle and headline, your connections count ("42 connections on xhuman", hidden below 3), Working on (up to 3 projects) and Open to (a service sits there with its price and Book), a few questions to ask your twin, and Ask, Follow and Connect at the bottom (only the doors you've opened). A stranger gets two free questions, then signs in with their phone and keeps talking. Further down: About, Background (Now, Before, Earlier: the lines you approve, else from your story), Knows about (your expertise) and your links — no wiki, no Services section, no score.
11. It's one page, the same for everyone: a stranger, a follower and someone you trust all see exactly the same thing, with no posts on it (your posts reach people in their summary and on each post's own page). Anything between the two of you goes through your xhumans: theirs asks yours for a time or passes you a message. Your twin can still answer each person from what you share with them. If you've made your page private, it opens only for people you've let in (one of your audiences or circles); everyone else sees that it's private. Nobody learns which of your circles they're in, or your phone number or email.
12. To check how people meet it, switch on Public view on /profile (the real page as anyone sees it, private sections hidden), or open your handle's link.
13. On your phone, the top of your hub has Share my link (the phone's own share sheet) and Show my code (a QR that fills the screen; they scan it, tap once, and you're connected).

Try asking

• Train my twin on this doc
• What have people asked my twin this week?
• Where do I change what strangers see about me?
• What does my page look like to a stranger?
• Share my link
• I'd rather not say what's hard for me — will you ask again?

Public profile

How to

1. Ask me to hide your page, take your profile offline, remove or take down your page, or make your profile private, and I switch Public profile off myself, at once (set_privacy). Ask me to make it public again to turn it back on.
2. Ask who can find you and I read both switches first (get_twin_status), so what I say matches your Settings.

Try asking

• Remove my public page
• Make my profile private
• Who can find me on xhuman?

Your profile page

How to

1. Open /profile for your page. The link and open-in-a-new-tab buttons at the top copy or open your public page. What xhuman knows about you is in Settings › Memory.
2. To change a section, tap Edit on it and finish the sentence in the line at the foot, or just tell me in chat what changed. Your page updates as soon as I've saved it.
3. To hide a section from the public page, switch it to Private; switch it back to Public any time. Public view shows it as a stranger would see it, so you can check.
4. To see what I know about you, open Settings › Memory. Each fact is a card that says where it came from; its ⋯ changes or forgets it (Forget has Undo for 5 minutes). “See exactly what xhuman reads” opens the real text I read on every turn, word for word; it never rebuilds anything by being looked at. A gap there is a gap in what I know: tell me and I'll fill it.
5. To try your xhuman as a stranger would meet it, switch on Public view and ask it something at the foot of the page.

Try asking

• How do I make a section of my profile private?
• Can I see my profile the way a stranger sees it?
• What does xhuman know about me?
• Show me exactly what you read about me
• Where do I edit my profile page?

Keeping your page true

How to

1. Ask “how do I look to people who don't know me?” — I tell you what strangers see on your page, what stays hidden (your number, email, circles, contacts, calendar, anything you didn't make public), anything that's out of date, and give you your link.
2. When your About or the line under your name is out of date, your evening note asks once: “Your About still says … Want it to say …?” Reply yes and it goes up; reply no (or nothing) and I don't ask again until something else changes.
3. A day after you join, if your About is still empty, I write a first draft of it (and the line under your name) from what you've made public with me, your email signature and your LinkedIn headline — never a job, link or claim I don't have. Your evening note shows it once, with your link last. Yes puts exactly that up; give me other words and I use yours; no and I never draft it again.
4. “Your page this week” ends your evening note on weeks people asked your xhuman about you: how many, and who asked what. Quiet weeks say nothing.
5. Change anything yourself by just saying it: “change my one line to …”, “my About should say …”.

Try asking

• How do I look to people who don't know me?
• What does my page say about me?
• Is anything on my page out of date?
• Who looked at my page this week?

Access

How to

1. Tell me, in chat or in the line at the foot of /profile ("I'm closed to recruiters", "only angel invest if I know the founder").
2. Turn on Public view on /profile and ask your xhuman as a stranger would, or ask me to try it as a recruiter or a founder raising who wants you as an investor, to see exactly what happens.
3. When your twin passes something to you, it arrives in your Twin queue and by email, Telegram and SMS.

Try asking

• Stop my twin from entertaining recruiters
• Only take investor pitches from founders I know
• Can I test how my twin answers an investor?

Knowledge base

How to

1. Ask a question about your knowledge base, or ask me to update an article.
2. Say “remember that…” — or “note that…”, “keep in mind…”, “don't forget…” — and I save it at once, from your own words (in that message, or the one just before it). In the app a “Saved to memory” card shows it with an Undo for 5 minutes; on Telegram, SMS, iMessage or WhatsApp I just tell you what I saved, and “forget that” within those 5 minutes takes it back.
3. Tell me when something I remember is wrong or has changed and I update it — the old version is kept as history; a new fact never quietly replaces an old one. “Forget …” takes a fact out of my memory at once: undoable for 5 minutes, then erased for good.
4. On your Memory page (/profile?facet=memory), a fact you take out there has an Undo for 5 minutes, and Export downloads everything I know about you as a file (Markdown or JSON) — facts stored encrypted come out readable in it, marked as encrypted, for you only.
5. I also pick up durable facts from what you type — and from a quick yes to my question about you — in the background. Everything I remember that fits (what you wrote yourself first) is with me in every conversation, here and on your messengers, plus anything new since; past that, the older notes closest to what you just typed come in too, each with the day I learned it — that's why I might bring up a trip you told me about months ago. The rest I can look up, and a fact you forget leaves at once.
6. To see what I know about you and where I learn each fact from, open your Memory page (/profile?facet=memory): grouped the way I read it — what you wrote, what holds for now, what I learned — and, where I have them, your own words and the chat and day you said them, linked. Add, edit or forget any fact there — what you write there is yours, and I never replace it on my own. A fact marked “Not used yet — waiting for your OK” came from something you didn't write, and I leave it out until you keep it: edit it and save it, as it is or corrected.
7. Ask me to bring your ChatGPT, Claude or another AI chat history in: if you're signed in to that site in my browser, I ask the site for your data export, check back when its email arrives, open the link and take the file in, then tell you how many chats came in and what they covered. If you aren't signed in there yet, I'll help you connect that site first. The two ways below still work too.
8. To bring your memory from ChatGPT, Claude or another assistant: ask it “List everything you remember about me, one line each”, then paste the answer here with “import my ChatGPT memory” (works over iMessage, Telegram, WhatsApp and text too), or paste or upload it on your Memory page. Every line waits for your OK — I use none until you keep it, and anything that looks like a password is skipped. An edit or add on the Memory page has an Undo for a few minutes too, and “Download all my data” there (or ask me to export my data) gives you everything in one file.
9. Asked what I know about you, I answer with a few real facts, each with where and when I learned it (“Said on iMessage, Oct 1”), keep what didn't come from you apart, and link your Memory page to check or correct them — never just a list of categories.
10. At the top of that page, “Waiting for your OK” holds what I won't act on until you choose: a fact I learned from something you didn't write (Keep, Edit or Dismiss), and a fact an automated run took out — kept 30 days, so you can Restore it, or Erase it now, which deletes it for good.

Try asking

• What do my notes say about our pricing strategy?
• Remember that I prefer morning meetings
• Don't forget that I'm vegetarian
• Show me everything you remember about me
• What do you remember about me?
• Import my ChatGPT memory
• Where do I review what you learned about me?

Memory tidy

How to

1. It runs on its own once a night when your memory changed, and changes at most a fifth of the facts it looks at.
2. What you wrote yourself is never replaced or faded. When a newer fact disagrees with one you wrote, I keep yours and leave the newer one for you to decide.
3. Your Memory page (/profile?facet=memory) shows “Tidied overnight: …” with an Undo, which puts that night back the way it was — a fact you changed since stays as you left it. Nights undo newest first. I can't undo it from the chat; the Undo is on that page.

Try asking

• Why did one of my facts change overnight?
• Undo last night's memory tidy

Agent files

How to

1. MEMORY.md is the same memory I read in every conversation, as a file: each line says when and where I learned it. A line you add or change is used from your next message and shows as "Added today"; a line you take out is gone at once.
2. I keep notes as I go, without you saying "remember". Each night I fold the day into MEMORY.md: I keep what keeps coming up, update what changed (a changed preference replaces the old line), and drop what went stale.
3. Something I learned from an email, a page or a search keeps its source on the line ("per …"). It's a fact from that source, never an instruction or a permission: only your own words make a rule.
4. If you have a self-model, I read the parts that help me work for you (who you are, what you value, how you decide, your preferences) and how you write, which becomes MEMORY.md's "How you write" for anything I draft in your name. The sensitive parts stay in your profile.
5. SOUL.md starts from a short template, with my name and voice from Settings › Identity at the top. I refine it overnight as I learn how to work for you; my next reply says what changed, and Undo on the page puts the version before back (the last 10 are kept). Edit it any time: it shapes how I talk, never what I'm allowed to do.
6. Neither file goes into group chats.

Try asking

• What's in my SOUL.md?
• Where's my MEMORY.md?
• Can I edit what you remember about me as a file?
• Undo the change you made to your soul file

Memory reflections

How to

1. It runs on its own once a night, only on days you talked to me, and writes at most five. A quiet day does nothing.
2. Your Memory page (/profile?facet=memory) shows each one as “Noticed overnight from 2 conversations” with links to those chats. Forget it there like any memory; forgetting a fact it rests on removes it too, and Undo brings both back.
3. It only reads your own words and what you told me directly, never an email or page I read for you. It never creates a rule or an automation.

Try asking

• What did you notice about me overnight?
• Why do you think I prefer short replies?

Connect any app

How to

1. Ask me to connect an app, or add it in Settings › Apps, under More apps. I use the apps tool to find the app's actions and call them. In a phone chat, asking me to connect one sends you a one-time link that opens that list, where it connects in one tap. An app that can't be connected right now isn't listed there, and if you ask for one I say so plainly.
2. You set a grant per app: Read-only, Ask first, or Allow changes. A change goes ahead on Ask first (the default) or Allow changes, while you're in the chat and when a job you set up runs without you.
3. While you're in the chat, a message, post, invite, share or campaign you ask for goes ahead without a card. A payment shows you its price once first, a delete asks once, and with “ask me before every send” on, every send asks. When you're not there, a message can still go out (unless “ask me before every send” is on); nothing else sends through an app.
4. Files move both ways between a connected drive and your computer: a file I fetch from Drive or Dropbox is saved to your computer, and a file on your computer (up to 3 MB) can be uploaded to a connected app. An upload you ask for in the chat goes ahead.
5. For an app that isn't in the catalog, I draft a connector from the API's docs. You approve it and paste its API key in Settings; the key is sealed under your own data key and I never see it. It can't be used before that.

Try asking

• Connect my Notion
• Can you post in my Slack?
• Put the report on my computer in my Dropbox
• Add a task to my Asana

Use me in Claude or ChatGPT

How to

1. To use xhuman, just talk to me here. I won't suggest adding me to Claude, ChatGPT or another app.
2. To have me use your Claude or ChatGPT account for you, press Connect on that row of the Finish setup page and sign in once.
3. If you set up a connector earlier, it keeps working. To cut it off, remove it under Old access in Settings › Apps or in that app's own settings.

Try asking

• Can I use you in ChatGPT?
• How do I add you to Claude?
• Do you work inside Claude?

Ideas

How to

1. By text or chat: “never tell me about crypto”, “only message me between 9am and 9:30pm”, “tell me about fundraising”, or “stop the daily ideas”.
2. Or Settings › How it works for you, under When it talks to you › Fine-tune by topic: “Tell me about”, “Never tell me about” and the Daily ideas switch. The hours I may start a conversation are the hours outside your quiet hours, set on the same page, unless you asked me for a narrower window.
3. A never-topic holds back anything I would start on my own that is about it, however it's worded; I never text you about it, and a reply I drafted for you on it shows as one line in your daily digest. The window holds my own messages outside those hours. I don't set location reminders yet.

Try asking

• Never tell me about crypto
• Only ping me between 9am and 9:30pm
• Stop sending me ideas

Buying things for you

Not available yet. This is not switched on for accounts today, so its steps are held back. This note clears itself when it is.

Allowed without asking

How to

1. By text or chat: “always allow emails to sam@acme.com”, “stop allowing emails to Sam”, or “stop asking me before inviting dana@x.com”.
2. On an approval card: tap This chat or Always (on a text, reply with the word Always).
3. Or Settings › How it works for you › What you've allowed › Allowed without asking, where each one has Take back.

Try asking

• Always allow emails to sam@acme.com
• Stop allowing emails to Sam
• Stop asking me before inviting dana@x.com

Devices

How to

1. There is no browser node extension to install and no xhuman app on an iPhone to pair: if someone asks how to install the extension or pair the xhuman app on their iPhone, say plainly it isn't available.
2. If someone asks to connect or pair their iPhone or Chrome, or to enter a pairing code, say it isn't available (a Mac, PC or Linux computer connects under Your own Mac or PC). For a site they are signed in to, ask me to log in to it and I send a secure sign-in link; the ones connected are listed in Settings › Vault.
3. Nothing is paired, so there is nothing to stop or turn off: I am not reading their photos, phone calendar or contacts. For anything on their phone, they can send it to me in chat (a photo, a screenshot, a forwarded message) and I work from that.

Try asking

• What can you do on my phone?
• Connect my iPhone.
• What devices are paired?
• How do I stop you from reading my photos?
• How do I use a site I am logged in to?

Profile & services

How to

1. Ask me to edit a section of your profile, or set up a service you offer.
2. I make the changes you ask for.

Try asking

• Update my profile's about section
• Set up a paid 1:1 advisory service

Your other email addresses

How to

1. IN CHAT: say "my other email is name@example.com", or "that's me" about a sender I mention. Say "remove name@example.com, that's not me any more" to take one off.
2. IN SETTINGS: open Settings, General, then Other email addresses. Add one, or press Remove on a row.
3. ONLY YOUR WORDS: an address I only saw in an email or on a page needs you to type it, so nobody can make me add one for you.
4. WHAT IT DOESN'T DO: it doesn't connect that mailbox or read its mail. To read it, connect it in Settings, Connections.

Try asking

• My other email is name@example.com
• That sender is me, it's my old work address
• Which email addresses do you know are mine?
• Remove my old address, that's not me any more

Connections & settings

How to

1. Ask me to connect a service; I show an inline connect card.
2. Ask what's connected and I list every connected account and whether each one is working, or update a setting like your timezone.
3. LinkedIn has a connect card like everything else — say “connect my LinkedIn” or “link my LinkedIn” and I show it. The LinkedIn card opens a dialog with two doors: live sync on any paid plan, or the free Connections.csv export that any plan can upload. Once your LinkedIn connections are in, I check them against your live searches that include your network and hold anyone who matches for your say-so.
4. Ask me to disconnect an account — “disconnect my Drive”, “remove my work Gmail” — and I revoke xhuman's access to it right there, the same Remove as on the Connected page; with more than one Gmail or Outlook connected, I ask which address.
5. Say “make it read-only” about any connected account — mail, calendar or a connected app — or ask for that when one is added, and I never send, draft or change anything in it, and I keep using what's there. Read-only on a connected account is enforced by us, not the provider: every write on it is refused on our side, so it never reaches them; the access you granted Google, Microsoft or the app at connect is unchanged. The provider's own screen still asks for the usual access when you connect. Say “allow changes again” in your own message and it applies at once.
6. Ask me to change your language — “speak English”, “set my language to Portuguese” — and I switch at once; it's also the Language row in Settings › General (English, Portuguese, French, German or Spanish). It is also the first thing a new person chooses, before anything else: a picker with the five languages by their own names, the browser's language preselected, and their onboarding cards and the text-consent line follow it. The navigation and many screens switch with it too; a few screens are still in English for now. It is the language I answer you in and write your goal titles, plans, path names and suggestions in, whatever language the email they came from was in; a goal whose title and plan were written in another language is rewritten in yours the next time you open Goals.
7. On a paid plan, when a connected account stops working — disconnected because its provider revoked or expired xhuman's access, or Gmail/Outlook stopped syncing new mail — you get a reconnect email (and a Telegram or SMS ping if linked) with a Reconnect button. It repeats at most three times per outage, three days apart, and stops once you reconnect — or once you choose Disconnect on that row, if you removed access on purpose. Open the Settings page it links to, or ask me to reconnect it and I show the connect card.

Try asking

• Connect my calendar
• What accounts do I have connected?
• Disconnect my Drive
• Make that account read-only
• Set my timezone to Pacific
• Set my language to English

Delete account

How to

1. To delete your account and everything in it, open https://xhuman.com/settings/general#delete-account and press Delete account, then confirm once. I can't delete the whole account from a chat, so I send you that link instead.
2. Deleting your account cancels your subscription first, so it never bills again, then deletes everything it holds: your profile, memories, chats, goals, saved logins, connected accounts, campaigns and network. It can't be undone or restored. You are signed out, and the account can never be signed into again.
3. It usually finishes in under a minute; if anything is still being removed, the page says so and it finishes on its own. The same email or phone number can sign up for a new, empty account afterwards.
4. To keep a copy first, ask me to export your data and I give you a download link.

Try asking

• Delete my account and everything in it
• How do I close my xhuman account?

What a disconnect keeps

How to

1. When you disconnect an account, my access to it stops at once: no new syncing, nothing sent from it, scheduled replies from that mailbox stop.
2. What I already read is not deleted right away by a disconnect. A disconnected mailbox's emails, its /mail conversations and the wiki pages built from its mail are hidden from /mail, email search, the wiki and my answers until you reconnect that same account, which brings them back as they were. A wiki page that drew on that mailbox is hidden whole, even when other sources fed it too. Hiding takes minutes for most mailboxes and can take a few hours for a very large one. The people and companies found in it, calendar events, imported files and notes, and our chat history stay as they are.
3. A disconnected Gmail or Outlook mailbox's data is permanently deleted 30 days after the disconnect, unless you reconnect that same account first. Ask me and I delete it right away instead, and tell you after what went — it cannot be undone. Our chats, your memories, your campaign records and the people in your network are kept. For other accounts, you can have me forget specific memories or articles, delete a source, a meeting or chats one by one, or delete your whole account.
4. Disconnecting an app such as Notion or Google Drive works differently from a mailbox: I tell you what is true for that account: how many items I pulled from it are still here, that I hadn't pulled anything from it when I can be sure of that, and otherwise that a disconnect deletes nothing I took from it. What I pulled from an app can be deleted in one step: ask, and I delete it and tell you what went. The imported items are kept out of sight for 30 days before they are gone for good, and there is no page to restore them from. A wiki page built only from them is deleted for good, and I say how many. A page edited since it was built stays.

Try asking

• If I disconnect Gmail, what happens to what you already read?
• Will disconnecting Outlook delete my emails?

Settings pages

How to

1. Ask where a setting lives (“link to my calendar settings”) and I give its full address. Your settings:
2. General (your name, handle, photo, phone, language, timezone, verification, who can see your profile and the communities you joined): https://xhuman.com/settings/general
3. Plan & usage (your plan, credits, usage, billing, invoices and payment; change plan and your wallet balance from there): https://xhuman.com/settings/billing
4. Identity (my name, tone, detail and the voice I use on calls, my email address, and Save to contacts): https://xhuman.com/settings/identity
5. Reach (texting (iMessage or SMS), Telegram, WhatsApp, browser notifications, my email address, and the words you can text me): https://xhuman.com/settings/reach
6. How it works for you (pausing everything; what I do on my own and what I ask you first (your email replies, websites, automations, meeting invites); money and purchase caps; how often I message you, quiet hours, the morning brief, meeting prep and which topics I tell you about; your always and never rules, and how you write email): https://xhuman.com/settings/how-it-works
7. Model (which model answers you in your chats (Sonnet, Opus, GPT and others) and what each costs): https://xhuman.com/settings/model
8. Files (your documents I can read and attach to email, which ones campaigns may send, and the answers campaigns may give): https://xhuman.com/settings/files
9. Email & calendar (your Gmail and Outlook accounts with their calendars (one row each, with what I read in each), signature (and an optional unsubscribe line at the end of campaign emails, off unless you turn it on), autocomplete, read receipts, new people per day, test sends, the spam test): https://xhuman.com/settings/email
10. Apps (GitHub, LinkedIn, Notion and other connected apps, connecting a new one, your Chrome extension, your team CRM, what else I read, and old access like AI apps and API tokens): https://xhuman.com/settings/connectors
11. Meetings (meeting recording and the notetaker, working hours, guest prep, booking links, paid sessions): https://xhuman.com/settings/meetings
12. Campaigns (sequence templates, tracking, saved search criteria, which campaigns send first): https://xhuman.com/settings/campaigns
13. If it isn't clear which settings they mean — connected calendars (Email & calendar) or their booking link and availability (the scheduling feature) — ask in one line.

Try asking

• Show me the link to my calendar settings page
• What's the link to my billing page?

Chat by Telegram, SMS or WhatsApp

How to

1. Turn a channel on in Settings → Reach (under Your agent). For texting on an iPhone, tap Open Messages (or scan the QR) and say hi from your verified number — the agent answers right away, sends its contact card (save it so it shows up with its name and photo), and the card flips to On by itself. On other phones, enable SMS. For Telegram, scan the QR.
2. Once on, message the bot / number like you message me here.
3. On iMessage, 👍 or ❤️ a confirm message to approve it and 👎 to cancel — or just reply in your own words. A tapback anywhere else is just a reaction.
4. On SMS these exact texts are commands: CLEAR (drop focus), STATUS (account + focus), PAUSE ALL (halt my work), HELP (this list), STOP (opt out), START (opt back in). Anything else is a normal question for me.
5. On Telegram, type / for the full command menu.
6. On the phone it's one continuous conversation — there's no fresh chat or reset to start. Move on to whatever's next and I follow along, keeping what matters from before.
7. Ask me to change what you're pinged about, or to set quiet hours — no command needed, just say it. In Settings › How it works for you, under When it talks to you, Fine-tune by topic has every switch by name, grouped by topic.
8. Quiet hours start on by default for a new account, 10pm–8am: what I start on my own waits until 8am, while results and anything you asked for (a task's result, a digest, a purchase or reply draft waiting on your approval) still arrive right away. Settings › How it works for you shows it under Quiet hours as the default; change it there and it becomes yours. Existing accounts keep what they had.
9. Want quiet for a while? Ask in your own words — "nothing tonight", "quiet for two hours", "quiet until I ask" — and I confirm in one line, then start nothing until it ends: alerts, heads-ups, daily ideas and results wait and arrive afterwards, and a brief due meanwhile is skipped. It is a one-off, separate from your nightly quiet hours: it lasts until the time you name, or until you end it — 14 days at most. Urgent email still comes through as on any other day, and so does my answer whenever you write to me. What you scheduled yourself (an automation, a campaign) keeps running. "Quiet off" ends it; so does End now at the top of Settings › How it works for you, which shows while it is on.
10. My own updates go to one app: the one you write me in most, the next as a backup. On the web they sit under From xhuman on Home.
11. Notification settings are account-wide: every switch (all alerts, event types, quiet hours) applies to all your linked channels, and there is no mute for just one in those settings. An automation or digest can name its own delivery channel ("just text me" for that one is fine). To stop one channel in the settings, unlink it on the same page where you turned it on (its own row, Unlink or Turn off). I won't unlink anything on my own.
12. An update I send first (not a reply) can go to your other linked app — Telegram or WhatsApp — instead of iMessage while that iMessage thread keeps its health. If iMessage is your only linked app, some of those updates are skipped until you text me again (your daily brief still comes by email); linking a second app means they still reach you. Texting me on iMessage works as usual.
13. How often I message you is the first choice under When it talks to you in Settings › How it works for you — or just tell me. Only when it needs me: approvals, replies you owe, money due and answers to what you asked, and I never start a conversation on my own (what I notice waits in the app). “Needs me, and what matters”, the default: also people you follow, goal progress and updates on my work, and my own messages only when they can't wait for your daily summary. Everything: every update and result, and I keep you posted whenever something moves. A choice sets the switches under Fine-tune by topic, on the same page; if you've changed some yourself it says Custom and what differs. It isn't a fixed number a day: I weigh each message against what you chose, what you already got today and what you act on, and what I hold shows up in your daily summary (with no messaging app linked there is no summary to hold for, so my own messages go at their old pace). It doesn't stop background work. Mute, on the same page, stops everything I'd send your chats on my own — and since I can't warn you then, a change your automations would make on their own waits on your list instead of going ahead; the morning brief has its own setting there (its hour, or Off).

Try asking

• How often do you reach out on your own?
• If I pick keep me posted, how often will you reach out on your own?
• How do I text you?
• What commands can I text you?
• How do I start a fresh conversation over text?
• Can I reset our chat on my phone?
• Set up SMS
• Can I text you on iMessage?
• Why don't your texts show up as blue bubbles?
• Can I approve things with a thumbs up?
• Can I approve with a tapback?
• Can I chat with you on Telegram?
• Stop pinging me about study alerts
• Why did my update come on Telegram instead of iMessage?
• Why do your updates come on one app?
• Stop pinging me so much
• Only text me — stop sending notifications to Telegram
• stop telegram notifications
• only text me

Messenger at signup

How to

1. After you sign in at xhuman.com/start, your app opens, and the first card above our chat shows how: on a computer, scan its code with your phone's camera (pick iPhone or WhatsApp) and send the message it writes; on a phone, tap the iMessage or WhatsApp button and send it. Or send the code shown to the number shown. The card says ✓ That's me in your Messages as soon as your message arrives.
2. Only a blue iMessage or WhatsApp proves the number is yours, so a green bubble links nothing: send it again as an iMessage, or use WhatsApp. If a message can't link your number (one from a country we don't serve yet, one already on another account), the card says why in one sentence, and so does my reply in that chat.
3. Telegram isn't offered at signup: add it, or another app, any time in Settings → Reach.
4. The phone app sends you to the web to finish signup until your chat is linked.
5. Not ready yet? You're already in the web app: the card just stays above our chat until your number is linked.
6. Need a card to start? If you've been invited, no: you start on Free (1,000 credits a month) at Launch, and the Pro trial is offered beside it. Only Pro's trial asks for a card.

Try asking

• Why did signup make me connect WhatsApp?
• Can I skip linking a messaging app?
• How do I finish signing up?
• Why don't I see WhatsApp when I sign up?
• Why don't I see Telegram when I sign up?
• Do I need a card to start?

Your app, and linking your phone

How to

1. On a computer, scan the code on that card with your phone's camera (pick iPhone, WhatsApp or Telegram): the message opens already written, so just send it, or on Telegram tap Start. Or send the code shown to the number shown.
2. On your mobile, tap the iMessage, WhatsApp or Telegram button on that card to link it: it opens with the message already written, so just send it, or tap Start.
3. The moment it arrives, the card says ✓ That's me in your Messages, and on your mobile Back to Messages takes you to our conversation there, where your setup goes on.
4. If it couldn't link your number, the card says why in one sentence. A green bubble can't prove the number is yours, for example, so send it again as a blue iMessage or use WhatsApp.
5. Wrote to me from a new number and got a sign-in link back? Sign in, and that card asks whether the number is yours: one tap is all it takes to link it, and I answer what you wrote.
6. Your xhuman, at the top of Home, shows the number to text me at, my own email address while it takes mail for you, and your web address, plus whether your phone is linked yet.
7. On the waitlist? Your app is open to you too: xhuman.com/waitlist opens your Home there.

Try asking

• How do I link my phone?
• Where do I text you from?
• What number do I text you at?
• What is Your xhuman on my Home?
• What's the card on top of my chat?

Telegram groups

How to

1. Add the xhuman bot to the Telegram group, from the Telegram account you linked in Settings → Reach.
2. Mention me (@ the bot) or reply to one of my messages — that's how I know a message is for me. Everyone in the group sees my answer.
3. The first time I speak there I say whose assistant I am and that I can see the chat.
4. Anyone in the group can mute me there with the bot's quiet command; only you can bring me back with its resume command.
5. If the group has none of my people in it, I leave.
6. When I pick something up with you privately (a booking, a send), I read what the group said myself — you don't repeat it — and when it's done I tell the group, if you asked me to.

Try asking

• Can you join my Telegram group?
• Can I add you to a group chat?
• How do I mute you in a Telegram group?
• Will you tell my friends my schedule in a group?
• Tell the group chat it's settled

iMessage groups

How to

1. Add the number you text me on — your own xhuman line, the one in our chat — to the iMessage group like any contact.
2. The first time you write in a new group, I say hello once and share my contact card there.
3. Saving that contact is optional: you only need it (call it xhuman) so your iPhone lets you @mention me.
4. Talk to me in plain words, mention me, or swipe to reply to one of my messages. Everyone in the group sees my answer.
5. I answer only you. Anything that books, buys or sends, I take to our own chat and do there with you.
6. Ask me in our own chat about what's been said in the group, even when nobody was talking to me, and I read it and tell you.
7. Groups with anyone on Android (green bubbles, SMS or RCS) aren't supported yet.

Try asking

• Can I add you to a group chat on iMessage?
• How do I add xhuman to my family group?
• Which number do I add to the group text?
• Can you join our iMessage group?
• What did Ana say in our group chat?

WhatsApp groups

How to

1. Message me in our own WhatsApp chat and I'll help there.
2. Ask me how to add me to a Telegram or iMessage group if you want me in a group today.

Try asking

• Does it work in WhatsApp groups?
• Can I add you to my WhatsApp group?

Agent identity

How to

1. Text me: “call yourself Jarvis”, “be more concise”, “be more casual”, “use the cedar voice”. I confirm what changed; “go back to xhuman” clears the name.
2. Or open Settings › Identity (under Your agent) to set the name, tone, detail and voice: a sample reply changes as you pick, and ▶ plays each one.
3. On iMessage my name shows once you save my contact card — Settings › Identity › Save to contacts, or Save contact under Text it in Settings › Reach (it needs you signed in on that device). The card carries my name, the xhuman mark, my email address and the number I text you from.
4. There's no picture to pick: my contact card always shows the xhuman mark.
5. The name and style apply to texts, the app and voice; email I send for you still goes out under your own name, and drafts I write for other people keep your own voice.

Try asking

• Call yourself Jarvis
• Can you be more concise?
• Can I change the voice you speak in?
• Which voices can you use on calls?

WhatsApp updates

How to

1. Link WhatsApp in Settings → Reach, then message me there once; updates start from that point.
2. Within 24 hours of your last WhatsApp message I send the full text. After that you'll see “Your xhuman has something for you: …” with a Show me button — tap it for the whole message.
3. Typing to me instead of tapping works too: whatever was waiting arrives first, then I answer you.
4. To stop WhatsApp messages entirely reply STOP; reply START to turn them back on. To mute updates on every app, tell me (notifications off).

Try asking

• Will you message me on WhatsApp when a task finishes?
• What is the Show me button on WhatsApp?
• Why did WhatsApp only send me a short update?

Stop everything

How to

1. Just ask me — or press it on Tasks, at the foot of the Goals list (/running opens it), or Pause everything in Settings › How it works for you. From your phone: /pause on Telegram or WhatsApp, or text PAUSE ALL (SMS or iMessage). Even mid-answer in a chat, I catch it right away.
2. It ends what's running now: browser tasks, any reply I'm in the middle of (in any chat — the one you asked in answers you with the report), code I'm running, background jobs, deep person searches, job scans and enrichment runs. Nothing after that point runs; a step already under way when you pressed it finishes. Messages from just before the press aren't acted on — repeat them if you still want them. Ended work doesn't come back with Restart: just say it again.
3. It pauses sending campaigns (and ones waiting to start after another), running searches and Scout, and holds the replies your inbox rules were about to send on their own — they wait for you, and a rule you agree to afterwards works as usual. Automations that send, draft or book pause, and so does any that runs me on a schedule (a digest, research); reminders and news watches keep going. Batches the product scheduled (like invites sent in bulk) and one-time tasks are held. Replies nobody reads wait for you as drafts.
4. It also switches off guest preps (no prep goes to your meeting guests, and a guest can't move a meeting from one), holds call invites waiting to go out from your calls, hands CC scheduling threads back to you (I book nothing more on them), pauses standing grants (and my scheduled suggestion runs) and auto-apply, keeps activity notes for your team's CRM waiting, stops the note-taker booking itself into new meetings, and refuses outside agents connected to your account (like Claude Desktop) when they try to send or change anything — their notes to your knowledge base still work.
5. Left alone on purpose, and listed in the report: meetings the note-taker is already booked into still happen, a job application already submitting finishes, reminder emails still go to people who booked a time with you, and an expert payment you already agreed is still released on schedule. Whatever was already being delivered when you pressed it is named as already on its way.
6. Nothing restarts by itself. Restart is one tap or word: Restart on the Goals tab (on its banner while things are stopped) or in Settings › How it works for you, the Restart button under the report on Telegram, /restart on WhatsApp, or texting RESTART. It restarts at once and says what goes out — how many emails (counted against your daily sending limits), call invites, prep emails and CRM notes, and what waits for tomorrow — and what was already due waits about 10 minutes, so there's still time to pause again. Replies it held and CC threads it handed you stay with you.
7. Texting STOP unsubscribes you from texts — it doesn't pause anything. Text PAUSE ALL, press Stop everything, or ask me instead. Turning texts off while things are still running gets you a note on your other channels. A YES never restarts anything — only RESTART does.
8. When you pause from one place, your other linked chats (and your browser, if notifications are on) get a short note that everything is paused. If it's still paused a day later, I mention it once in your next roundup.

Try asking

• Stop everything you're doing
• Pause all my stuff
• What did you stop?
• How do I turn everything back on?
• If I text STOP does it stop the agent?
• Does it hold guest preps and call invites?

Stop one thing

How to

1. Just say what to stop ("stop the follow-ups on the Acme thread", "cancel that browser task", "stop enriching @acme", "end the @acme campaign for good"). I stop that one thing and tell you exactly what stopped — or that it had already finished, or that I couldn't confirm it.
2. Ending a campaign for good means none of the emails it hadn't sent yet go out, first emails and follow-ups alike, and it can't be resumed; replies to what already went still come to you. If you might want it back, I pause it instead. On Tasks, the Stop on a campaign or a running job does the same right there.
3. A stopped reply waits for you as a draft; a stopped column keeps the cells it already filled; a batch already under way finishes its current step. Something that already went out can't be recalled.
4. To stop everything at once, use Stop everything instead.

Try asking

• Stop the follow-ups on the Acme thread
• Cancel that browser task
• Stop enriching the campaign
• Leave the meeting

Images, GIFs, video, voice and podcasts

How to

1. Just ask in your chat: “make a poster that says OPEN MIC FRIDAY”, “an 8-second video of waves at sunset”, “read this out loud in Adam's voice”, “turn this article into a podcast”. Voices: Rachel, Adam, Bella, Josh. A video takes about two minutes; if it isn't ready when I answer it arrives in the chat on its own.
2. To build an image from a picture: send me the photo and say what to make — “put my dog at a birthday party”. An image of something that is yours — your dog, your home, your car — is yours only when it is built from a picture of it: with none, I ask for one photo first (or, if you'd rather skip it, I draw a likeness and tell you it is one, not your dog). To edit an image I made, just say what should be different — “make the hat blue” — and the rest stays the same.
3. A GIF: just ask, “send me a gif of a dancing platypus”. It's about a second of motion that loops forever, drawn from your words, so I can't turn a photo into one. On a really big day of yours I may send one on my own, at most once a week and at no cost to you. Tell me you'd rather not and I'll stop.
4. I look at every image before I send it. If it missed what you asked — a word spelled wrong, a hat hiding a letter — I fix it once myself, at no cost to you, and send only the fixed one.
5. I won't make sexual content, a real person's face or voice (no deepfakes, no cloning), or anything meant to pass as someone else — so I can't build an image from a photo that shows a person's face (send one with just the pet, the object or the place), and I don't edit a picture of an ID, a receipt or another official document. I can make an original character or use one of my own voices instead.

Try asking

• Make me an image of a poster that says Open Mic Friday
• Make me a gif of a dancing platypus
• Can you make a short video?
• Turn this article into a podcast
• Read this out loud for me
• Can you generate images?
• Make a birthday picture of my dog from this photo
• Can you edit the image you just made?

Browser notifications

How to

1. Turn it on in Settings → Reach → Browser notifications: click “Turn on for this browser” and allow notifications when your browser asks. Do it once in each browser you want notified — each one you've turned on is listed there, with when a notification last showed on it, a test button and Remove.
2. On iPhone and iPad, notifications only work once xhuman is on your Home Screen: tap Share, then Add to Home Screen, open xhuman from there and turn them on in Settings → Reach.
3. They follow the same switches as Telegram and texting: proactive notifications on or off, each kind of alert on its own, and quiet hours (what arrives during quiet hours waits until they end).
4. Browsers get a short notification that links to the details — never a whole reply. When you have no messenger linked and no browser turned on, results of tasks you started come to your email instead. They come to your email as well when none of your browsers has shown a notification in the past week, or its notifications keep failing — so a laptop left closed never swallows one.

Try asking

• Can you notify me in my browser?
• Turn on browser notifications
• Send me a desktop notification when it's done
• How do I get notifications on my laptop?
• Why am I not getting notifications on my iPhone?

Threaded updates

How to

1. Nothing to turn on: ask for something recurring in your chat (“tell me tech news every morning”) and each update lands under that message.
2. Swipe to reply on any update to change THAT request: “less of this”, “stop”, “move it to evenings” — I know which rule it is.
3. Plain SMS (green, Twilio) has no reply threading; updates there arrive as separate texts.

Try asking

• Why are my news updates under my first message?
• How do I stop one of these updates?
• Can I reply to an update to change it?
• Why don't my texts thread like iMessage?

The 👀 on your message

How to

1. Nothing to do: you don't need to reply or send it again. The answer lands in the same chat as one message.
2. You'll only see one reaction from me on a message. A ❤️ on a thanks, or a 👍 when I've done what you asked and there's nothing to add, replaces the 👀.
3. It only goes on something you sent me (a text, a photo, a voice note), never on a button you tapped.

Try asking

• Why did you put eyes on my message?
• What does the 👀 mean?
• Why did you react to my text instead of answering?
• Why didn't you answer right away?

Answering on your phone

How to

1. To answer a question in a check-in I sent while you were away, 👍 it (or ❤️ it); 👎 means no. A 👍 on a message that didn't ask anything is just a 👍.
2. To answer an older question or card, swipe to reply on it and type your answer — it goes to that one, even if I asked something newer since.
3. Or just tell me in words — 'ok send', 'send the Mima one', 'drop it', 'don't send that reply' — on iMessage, WhatsApp or Telegram, and I'll do exactly what the card or notice you saw said: a draft, an intro, a question I asked, an inbox reply that waits or goes out soon; if more than one is open, or I asked you something else since, I'll check which. On a plain text line (SMS) words can't prove it's you, so a send needs the card in xhuman there (a 'no' still works by text).
4. When I pause a task to ask before pressing a button, a 👍 or a quick reply lets it go ahead.
5. Telegram has buttons under each question, so tap those there.

Try asking

• Can I thumbs-up your question instead of typing?
• Which card does my yes answer?
• How do I answer an older question on my phone?
• Does a 👍 approve things?
• Can I just type 'send it' to send your draft?

Campaign results on your phone

How to

1. Start the campaign from the chat you want to hear back in: its results come to that chat only, not to every chat you have linked (if that chat can't take a message, your next chat gets it).
2. Approve the reply draft from your phone with a thumbs up: a 👍 on the message (iMessage, WhatsApp), Telegram's Send button, or typing “yes” sends exactly the draft shown; “no” leaves the draft unsent in the campaign.
3. Want the draft changed first? Say how (“shorter, then go”) and I rewrite it.
4. A follow-up draft — someone said they'd book a time and didn't — reaches your phone the same way.
5. Results that need nothing from you don't text you one by one: they're a single line in the next morning's digest. In quiet hours the rest wait, then arrive one by one.

Try asking

• Why don't I get texts when someone replies to my campaign anymore?
• Why did the campaign reply texts stop?
• How do I send a campaign reply from my phone?
• Can I approve a reply draft with a thumbs up?
• Can I approve the reply with a thumbs up on my phone?
• Approve a campaign reply draft from my phone
• Will you tell me when someone replies to my campaign?
• Let me know when a meeting gets booked from my campaign

Invites & referrals

How to

1. Ask me for a link (say who it's for), or open /invite: each of your 3 invites is a card that says Ready, Shared, Tied to a number, or Used by the person who joined, and Share hands out the next one.
2. Rather not share a link? Give me their phone number and I tie one of your 3 to it — when they sign up at xhuman.com/start and text me from that number, they're straight in. We never text them.
3. Each person who joins with one of yours, gets set up and keeps texting me earns you one back, never more than 3 unused at a time.
4. You earn credits when someone you invited subscribes to a paid plan.
5. Used all 3? Tap Ask for more on /invite (free members can also text me MORE INVITES).
6. "Invites" on its own means these xhuman ones; LinkedIn's daily and weekly limits are answered only when LinkedIn is named.
7. Unless you pay for your plan by card, a verified phone has to be on your account before you can hand out invite links or send team invites, and before someone you invite to connect skips the waitlist. A phone you text me from (text, iMessage or WhatsApp) already counts; otherwise add your number in Settings › General and your invites work right away.
8. Paid and early members: your vouch link (xhuman.com/i/your-name) moves the people who open it up the waitlist. On /invite it's under Share, More options, with who has joined and the people you're closest to who aren't on xhuman yet.

Try asking

• A link for my friend Sam to join
• Tie an invite to Sam's number
• How do I get my friend in?
• Invite Maya, her email is maya@northwind.example
• How many invites do I have left?
• Can I get more invites?
• What's my invite code?
• Where's my vouch link?
• Do I get credits when someone I invite subscribes?

Getting in: waitlist, invites and phone sign-in

How to

1. Join or sign in at xhuman.com/start: your name, then Continue with Google or your email (I email you a way in). A new account then texts me once from your mobile, and we're set. There is no password to remember.
2. An account made with a mobile number signs in the same way it always did, from the small option under that form; an account with a password uses "Sign in with it" there.
3. Opened a member's invite? Sign in and you skip the waitlist and the plans.
4. See where things stand on your Home in the app (xhuman.com/waitlist opens it), or text STATUS. Text UP for the two ways in sooner.
5. Two ways in sooner: know a member and ask them to invite you (they just tell their agent), and connect more, because I'm most useful across your whole network and the more you connect, the better your chances of an early invite. We never promise a date or a number of days.
6. A member's invite link gets you in directly — members get 3, each good for one person. A plan gets you in now as well: ask me in our chat and I send the payment link, which shows the price and brings you back to the chat (or the Pro link on /waitlist).
7. Right after you sign in, your app opens, and the first card above our chat is how you text me from your mobile: on a computer, scan its code (pick iPhone or WhatsApp); on your mobile, tap the iMessage or WhatsApp button. It opens with the message already written, code and all — just press send and we're linked.
8. Want me in a different app too? Open xhuman.com/start?s=channel (members: "Add another app" on the last screen) and pick WhatsApp, Telegram or iMessage/Messages — I answer on every app you link.
9. If the app didn't open, or it opened on the wrong device, tap "Use a code instead" on that screen and text me the code shown (WhatsApp and iMessage/Messages).
10. Picked an app but never sent the text? I send one reminder, once — a text on Android Messages (never at night) or an email with a link that reopens the app you picked.
11. The sign-in links I send in chat work once, for an hour. If yours expired or was already used, text LINK and I'll send a fresh one.
12. Whatever you open from our chat in a browser, a sign-in, a "you're connected" screen, even an old address that no longer works, ends with one tap back to the chat it came from ("Back to Messages", Telegram or WhatsApp). Nothing there is lost: ask me again in the chat.
13. Text IDEAS in our first chat and I list the things you can just ask me for (a booking, a purchase, a trip, a price to watch). It only lists what your account can do right now. It is not the daily Ideas, which are grounded in your mail, calendar and goals.
14. Lost your phone, signed in on a shared computer, or someone else has your chat? Settings → General (your profile) → Sign out everywhere (accounts made with a phone number or with "Continue with email") signs out every browser signed in to your account, this one included, and disconnects apps you connected with a key or a sign-in, and any chat that isn't on your verified number, like Telegram (reconnect them after). Texting and WhatsApp on your verified number keep working.
15. If the iMessage number I text you from is taken out of service, you get ONE email from us with a tap-to-text link to my new number. Tap it and say hi, and we pick up there; nothing else changes.

Try asking

• Am I still on the waitlist?
• When do I get off the waitlist?
• How do I move up?
• How do I move up the list?
• Where am I on the waitlist?
• How do I skip the line?
• How do I log in without a password?
• Is my profile public?
• Can you text me on WhatsApp instead?
• How do I move my agent to WhatsApp?
• The app didn't open — can I use a code?
• My link expired
• How do I sign out of all my devices?
• Why did you email me a new number to text?
• What does the IDEAS text do?

Waitlist approvals

How to

1. Reply to the morning list with the numbers of the people to let in, like 1 2 or 1, 3, or ALL for everyone on it. Reply from iMessage, Telegram or WhatsApp — plain SMS can't approve, since a text can be sent from any number.
2. The list takes typed numbers while it is the last thing I asked you: once I've sent you another question or card, a bare number answers that instead, and a reply swiped onto some other message is about that message. More numbers later the same day still work while the list is the latest thing I asked.
3. Each person gets their welcome three days after they finished onboarding, or the moment you approve if that's later — never before, and only between 9am and 8pm their time. My reply says who and which day.
4. Putting someone on hold clears their approval; lift the hold and they're back on the next morning's list.
5. Anyone you don't pick stays on the list and shows up again tomorrow. The admin Waitlist tab can still let someone in by hand.

Try asking

• How do I let people off the waitlist?
• How do I approve people from the waitlist?
• What's the daily list?
• How do I let everyone on today's list in?

xhuman on the web: text LINK

How to

1. No chat yet? You can still use xhuman.com: on “Text xhuman to get started”, tap “You can also use xhuman on the web, anytime”. Until you add a chat, what I'd text you comes by email; add one anytime in Settings → Reach. Still on the waitlist? That line reads “See your waitlist” instead and opens your waitlist page, with plans one tap away.
2. On a free account, text LINK in our chat, then tap what I send back on your computer: you're in, no password.
3. Made your account with Google? The tap goes through Google and brings you back where you were going.
4. Each one is good for an hour: the one-tap kind works once, and the Google kind can be tapped again within that hour, since Google does the signing in. Text LINK again for a fresh one.
5. Inside Instagram's or Facebook's built-in browser, Google won't run: use Safari or Chrome.
6. On a paid plan, sign in at xhuman.com/start with your number, or at xhuman.com/sign-in with your email or Google, whichever you made the account with.

Try asking

• How do I get to xhuman on my computer?
• Can I use xhuman on the web?
• Can I use xhuman on the web before I set up a chat?
• Send me a sign-in link
• Text LINK

Back to your chat, and link previews

How to

1. Tap “Back to Messages” (or Telegram, or WhatsApp): at the bottom of a done page like “Card saved” or “Connected”, at the top of a page in the app. It opens this same conversation, and what comes next arrives here.
2. No button on the page (an older link, or one opened somewhere else)? Just switch back to your messages app: our chat is where you left it.
3. In the chat, my links show what they are — “Your meeting notes”, “Connect your calendar”, “Talk to xhuman” — never your name or what they're about, so a forwarded link gives nothing away.

Try asking

• How do I get back to our chat?
• I opened your link, how do I go back to Messages?

Credits

How to

1. What you type to me costs a small fraction of a credit per message on a paid plan. An automation that runs me works on its own, so it pays the full rate — usually tens of credits a run — from your weekly allowance first.
2. A browser task uses credits for the computer work it does — a few for a typical task, more for a long one — drawn from your weekly allowance first, then monthly, then bought. A code run does too, for the AI and PDF work it asks for and its time in the sandbox — usually about a credit, drawn the same way.
3. Your monthly credits are used first, then credits you bought; work I do on my own draws on this week's allowance first. See this week's allowance and your monthly and bought credits in Settings › Plan & usage, where you can top up or turn on Auto-reload so you never run out. The Usage page shows what used them.
4. Prices, in credits, when you ask: 25 for each person a search confirms as a match, however the search found them (55 instead, when Scout's search depth is Both sources, for each person the second source returns — even one the first found too); 7 for a work email and 22 for a phone number, each only when one is found, and 1 per address for a fresh email check; 10 for each person's profile on a list, 2 for a network contact's profile (once), and 10 to look up the work history of someone you email beyond the closest contacts covered free; 8 for each web lookup about a person or company, up to 26 for a web check on a network contact, and 6 to find a company's owner when its own site doesn't name one; 350 for an opportunity scan and 1 to score one job's fit; up to 14 for a person deep search, while a saved one costs what its first run did, at least 8; 84 for a spam test; 200 for an introduction through the network, once it's emailed; an image 6, speech about 18 per 1,000 characters, a podcast up to 67 and an 8-second video 180. Chat, an automation run, a browser task, a code run and a phone call are priced from what they actually use.
5. You earn credits when someone you invited subscribes to a paid plan — ask me for your link, or see Invites & referrals.

Try asking

• How many credits do I have left this week?
• Why did my automation use credits?
• How many credits does a browser task use?
• What happens when my weekly credits run out?
• What's the weekly allowance on Pro?

What's running and what it used

How to

1. On the web, open Goals and, under In the background, What I'm doing for you, then pick a tab: All, Coming up (what's due next, by day), Always on or Paused. It lists your goals, one row for the follow-ups from your mail (a count of each kind), and your watches, rules, waits, checks and reminders, paused ones included, each in your own words where you gave them; a row that has a page opens there so you can change, pause or stop it.
2. Ask “what's running?” or “what did each thing cost this week?” — I list every automation, check-back and watch with its state, its last run and its runs in the last 7 days.
3. Anything you cancelled or removed, or that has ended, is said as exactly that, with when — never as still running.
4. What each used is in the credits actually charged to you, the same charges the Usage page counts; your own messages and work I started on my own are counted as a whole, not per item.
5. I also say how much of your plan's allowance you've used and when it resets.
6. Want a heads-up when spending runs high? Ask for a rule like “every evening, tell me if this week passed half my allowance” — its own runs are charged too.

Try asking

• What's running and what did each cost this week?
• What do I have running right now?
• What are you doing for me?
• What did my automations use this week?
• How much of my allowance have I used?

Plans

How to

1. Ask me “what plans do you have?” or “how much is Pro?” in the chat, and I'll tell you each plan's price and what it includes.
2. Pro starts with a free trial: a card up front, and nothing is charged if you cancel before the trial ends.
3. Teams puts your company on one plan, with one shared pool of credits for everyone on it. To set it up, open xhuman.com/start/team.
4. Free: invite-only — join the waitlist, and connect your accounts to get in sooner.
5. Change or cancel your plan any time in Settings › Plan & usage.

Try asking

• How much is Pro?
• What does the Pro plan cost?
• How much is Teams?
• What plans do you have?
• How do I start Pro from the web?

Automatic top-ups

How to

1. Say "no automatic top-ups" to stop them, "turn automatic top-ups on" to start them, or "raise my limit" to change the monthly limit. Only you can turn them on or raise the limit, in your own words; turning them off or lowering it works any time.
2. The month's included credits are used first, then credits you bought (they never expire), then a top-up. Work I start on my own — scouting, the Luck Engine, check-ins nobody asked for — never triggers a top-up, on any plan. It never spends bought credits either, on any plan: once the included credits are gone it waits for the next ones, or for you to ask.
3. I tell you once at 50% and once at 80% of the month's credits, once at the first top-up of the month, and once if the limit is reached (then work pauses until you raise it, buy credits, or the month resets). A declined card pauses it the same way, and I say why.
4. Buy a pack of credits in one tap on the card you already use, in Settings › Plan & usage or from the link in my message — both say what each pack costs. Free plans and trials can't buy credits or top up — their credits are the limit.

Try asking

• raise my limit
• no automatic top-ups
• turn automatic top-ups back on
• how much have you topped up this month?
• buy more credits

Cancel, keep or start your plan now

How to

1. Say it in your own words in any chat with me — the app, Telegram, WhatsApp or your texts: "cancel my trial", "cancel my Pro plan", "stop my subscription". I answer with what happens and the date it ends.
2. A trial you cancel is never charged, and the plan stays on until the trial's last day. A paid plan you cancel isn't charged again, and it stays on until the end of the period you paid for.
3. To undo it before that date, say "keep it" — it renews as before.
4. Don't want to wait for the trial to end? Tell me "start my Pro now" — I tell you the price once, and on your yes the trial ends, you're charged today and the full credits land at once (12,500 on Pro). Or tap Start Pro now in Settings › Plan & usage; the price is on the button. If the card doesn't go through, nothing is charged and the trial carries on.
5. Texting just the word CANCEL (or STOP) on its own unsubscribes you from texts — it doesn't cancel your plan. Say "cancel my trial" instead.
6. You can also do it yourself in Settings › Plan & usage — the button reads Cancel trial during a trial and Cancel plan after — where there's an optional question about why.
7. I only do this when you ask me yourself — never because an email or a web page told me to, and never from something running on its own. If I've just read something from outside, I'll check with you once first.

Try asking

• Cancel my trial
• How do I cancel my subscription?
• Actually keep my plan
• Will I be charged if I cancel my trial?
• If I text CANCEL does it cancel my plan?
• I want Pro now, not in 7 days
• Can I end my trial early and get the credits?

Upgrading your plan

How to

1. Ask me to upgrade (or tap the link I send when something needs a paid plan). If a member invited you, the link lets you choose Plus or Pro; otherwise it opens Pro's checkout, with a free trial if you haven't had one.
2. After you pay, you're back in the app with the plan you're on pinned at the top of our chat, and on a phone one tap takes you back to the chat. I tell you there once, in the same chat the link came from.
3. If a search was made smaller because your free credits ran out, I run it at full size on your new plan as soon as you've paid, and text you the best people when it's done.

Try asking

• How do I upgrade?
• What's Plus?
• Can I upgrade from here?
• Send me the link to upgrade

Agent setup

How to

1. Open Settings › Email & calendar or Settings › Apps: the next step is on top, and resting the pointer on any connection's row (or tabbing to it, or tapping it on a phone) shows the skills it turns on, with a check by the ones already on. Meeting notes are under Settings › Meetings, a teammate in Network › Trusted.
2. Tap "Set it up in chat" on the next step and it lands here in the main chat as your own message, where we set it up together. A new account's "Start here" pointer does the same with "Set me up".
3. Or just ask me here: "what should I set up next?", "connect my calendar", "add a teammate".

Try asking

• What should I set up next?
• Show my setup progress
• What am I not using yet?
• What does connecting my CRM unlock?
• Where do I see what each connection turns on?

Finish setup (optional extras)

How to

1. On your laptop or desktop, use the Finish setup card on Home, or open the Finish setup link if I texted you one.
2. Claude or ChatGPT: press Connect, then sign in once in the window that opens. After that I can work in your account for you in my own browser. Signing in on the page does not bring in your past chats; the page also lets you bring in an export file of past chats if you have one.
3. Or just ask me in the chat (“connect my Claude so you know my past chats”, “connect my ChatGPT”): I send you the same sign-in for claude.ai or chatgpt.com right there, and once you're signed in from the chat I start bringing your past chats in. It is never a sign-in to Claude Code or Codex on a computer; that is only for a job you ask to run on your plan.
4. Your computer: on a Mac, press Download for Mac (just Download on the Home card), install the app, then connect it right there; elsewhere press Connect for the one-time link and the line to paste. Once it is connected I can work on it from the chat, including running Claude Code or Codex there.
5. Email and calendar: press Connect and pick Gmail or Outlook. With Gmail the row then offers Google Calendar as the next step; Outlook's connection covers its calendar.
6. Press Not now on any of them (the × on the Home card) and it comes back on the card in a week. Opened on a phone, the page says this part is done on a laptop or desktop, with one tap back to your chat.

Try asking

• Where do I finish setting up?
• What's on the finish setup page?

Web search

How to

1. Just ask a factual question and I look it up.
2. Point me at a chart, a screenshot or a scan (a link to the image, or a chart my computer drew) and I look at the image itself.

Try asking

• What's the latest funding round for Anthropic?
• What's the exchange rate for EUR to USD?
• Look at this chart image and tell me the October number
• What does the screenshot in that link show?
• What is the trend in this line chart?

Places nearby

How to

1. Name the spot and what you want (“a quiet cafe near Union Square”, “the closest pharmacy to 5th and Main that's open now”) and I list what's around it, nearest first, from live Google Maps listings.
2. Distances are straight lines measured in code — from the edge of a park or building when its outline is known, so “within 200 m of the park” means from the park, not its center. I say it's straight-line when you ask about walking or driving.
3. I check every condition you give (open at 11pm Wednesday, wheelchair accessible, vegetarian, rating) against each place's own listing, and rule places out for a stated reason. Hours are the listing's regular week; a holiday can differ.
4. If you don't say where, I ask — I never guess where you are.

Try asking

• What's the closest pharmacy to me that's open now?
• Find a quiet cafe near Union Square
• How far is the gym from the park?

Product search

How to

1. Say what you want in your own words, with anything it must have (“a black silicone band that fits my Garmin Approach S72, under $40”), and I come back with up to ten products.
2. Each one shows its photo, the price, the store and whether it's in stock, read from the store's page. If a page doesn't say whether it's in stock or ships to you, I tell you it didn't say.
3. I look in your country and show prices in your currency; a store that only lists another currency is left out rather than converted at a rate nobody stated.
4. I check which ones actually fit what you asked and recommend one — then you pick, and I can take it from there.

Try asking

• Find me a band for my Garmin Approach S72
• A burr coffee grinder under $150
• Show me clear cases for an iPhone 17

Food and calorie log

How to

1. Send a photo of your meal, a nutrition label, or a few words, and I log it in one short reply with the calories and protein.
2. When one guess would change the number a lot (how much dressing, a small or large), I log my best guess and give you a one-tap way to change it. I only ask outright when I can't tell what the food is.
3. Correct me any time (“it was a large”, “I didn't eat the fries”), or say “delete that” — a removed meal can be put back within a day.
4. Ask “how many calories have I had today?” and I add up what you logged, against your targets if you've told me them; I say when I only know some of the foods' nutrients.
5. If you already keep food on one of your pages, I link it to your food log so it shows every day's totals and meals on its own — one record, nothing to type into it.
6. Say “stop asking me” and I just log my best guess.

Try asking

• Log my lunch
• How many calories have I had today?
• Two eggs and toast for breakfast

Weather forecast

How to

1. Ask “what's the weather this week” and I read the forecast for the city saved on your profile, or your one ongoing stay on your Explored map. Name a city (“weather in Lisbon”) and I read that one.
2. I never guess where you are: if no city is saved, or two stays are ongoing, I say there is no location instead of picking a place. Saving a city on your profile fixes it.
3. A recurring rule (“every morning, text me the weather and my day”) uses the same read. A rule's run reads a named city only when the rule's own words name it, or it is your saved city.
4. Temperatures come in °C and °F, rain in mm and inches, wind in m/s and mph. The days are cut in your own time zone. Data from MET Norway (CC BY 4.0), cached for about an hour.

Try asking

• What's the weather this week?
• Will it rain in Lisbon on Saturday?
• Include the weather in my morning rule

Home panels

How to

1. Just ask: “follow Tesla stock”, “put my next meeting on Home with a Join button”, “count down to my board meeting”, “show the people I follow”, “track the seed round”, “weather in Lisbon on my Home”. Or tap “Create a new panel” at the foot of your panels on Home and type what to track there. The panel shows up on Home right away while it builds; a stock, calendar, follows, goal or weather panel goes live within seconds.
2. For anything that needs me to think it through, ask for it the same way (“a panel with the AI news that matters each morning”). I write its lines now and refresh them once each morning.
3. Each panel has a small menu: Refresh, Edit in chat and Remove. You can also say “remove my Tesla panel” or “move the weather to the top”.
4. Stock, calendar, follows, goal and weather panels refresh hourly and cost nothing. A panel I write uses about one of my turns a day from your credits, and you can have up to 3 of those (12 panels in all).
5. Prices come from Yahoo Finance's public feed and may be delayed; the weather is from MET Norway.

Try asking

• Follow Tesla stock on my Home
• Show my next meeting with a Join button
• Add a panel with the people I follow
• Track the seed round on Home

Flights

How to

1. Tell me the trip (“NYC to Chicago Friday morning, back Sunday evening, under $400”) and I search real offers, cheapest first, with the airline, flight numbers, times, stops, fare brand, bags, and whether the fare can be changed or refunded and for what fee.
2. Ask for the cheapest (“the cheapest one-way out of New York anywhere in the US this week”) and I compare every airline Google Flights lists, from every airport of the city, on every day you allow and to as many places as the ask leaves open, then give you my pick, one runner-up and what I compared.
3. Ask for a seat (“an aisle seat”) and I read that flight's seat map: which seats are open, each one's price, and the total with the seats added — the fare as it stands now, not the price from the first search.
4. Aisle, window and middle come from the cabin's layout, not the seat letter. When the layout around a seat can't tell, I say so instead of guessing.
5. Offers expire, usually within the hour, and I say when. When you pick one, I show the flights, seats and total in one line. The first time, your yes gets a link to pay on our secure page — signed in as you, the same total on the Pay button, nothing charged until you press it — where you can tick “Save this card so I can book flights you OK in chat”; leave it unticked and the card isn't kept.
6. Once that card is saved, the line names it (“$412.80 total on your Visa ending 4242. Book it?”) and your yes books it right there in chat, charged exactly that total. If the price changes before I book, nothing is charged and I ask again with the new total; if your bank wants to confirm the payment, I send you a link to approve it. Where booking here isn't on, I book it on the airline's own site with your yes.
7. Paying with points or miles (“use my points for business to Lisbon”, “how many Aeroplan miles is it?”): I look for award seats on a points-search site that needs no sign-in and on the airline's own award search, and, signed in to your own airline or loyalty account (your saved login, or you connect it once), I read your balance and the award space only members see. You get each award option in points plus taxes and fees beside the cash fares, and I tell you which ones needed your sign-in.
8. To book an award seat, I tell you the points it spends and the taxes and fees in cash once; when you agree, that covers the whole job, signing in included, and I book it on the program's site with your account.
9. To stop a chat yes charging your saved card, turn on “ask me before every purchase”: every flight then asks you on our secure page first. Removing a saved flights card is a support request for now.

Try asking

• Find me a round trip from New York to Chicago Friday morning, back Sunday evening
• What would an aisle seat cost on the cheapest one?
• Is that fare refundable?
• What's the cheapest one-way out of New York anywhere in the US the next few days?
• How many points for business class from New York to Lisbon, compared to paying cash?

Passport and ID details

How to

1. Say where it is — “my passport is in my Drive” — and I find it there by what it is, no file name needed: your own pictures and PDFs that read like an identity document first, then your newest ones, and the first whose coded lines check out is the one (at most three are read). Name the file and I look for that name; if several match, I ask you which. A file already on your computer works too.
2. I use only the two or three coded lines of letters, digits and < at the bottom of the photo page (the back of an ID card), and every check digit in them has to match. A blurry, cut-off or glared scan is turned away, never guessed at.
3. When a job needs your passport and your vault doesn't have it yet, I ask you once, right in the chat, for just the details the site's form wants: you can type them, or send a photo of the photo page or say where it is in your Drive if that's easier. Then I carry on with the job in the same reply.
4. What I take from your own file is kept, sealed under your own key, in your vault until you delete it — ask me in the chat, “delete my passport from the vault”, and it's gone for good — and the next job uses it without asking. A file someone shared with you is kept for 24 hours only. The file itself stays where it was — in your Drive or your files — and a photo sent in a chat stays in that conversation like any picture you send.
5. When you ask for something a site needs it for — an airline's online check-in, a hotel registration — or a job you set up reaches that site, the details are typed by code into that one site's own page. They never go into my messages, and the browser's report has them taken out.

Try asking

• My passport is in my Drive — when does it expire?
• Use the passport scan in my Drive
• Is my ID card still valid?
• Here's a photo of my passport for the flight
• Delete my passport from the vault

Browser tasks

How to

1. Tell me the website and what you want from it (“check their careers page at acme.com/careers and tell me which roles are open in London”, or “open this link and summarize the page”). A task also starts on its own from an automation, a task you handed me, a helper or a check you asked for, acting on what your own words for that work asked (see “Automation browsing” for how). Every task you ask for starts right away, with nothing to approve first. One that will buy asks you once, with the price, before it pays — unless you already said yes to that price in our conversation; a send, post or sign-up you asked for goes ahead with no second question, unless you told me to always ask first.
2. A task that only looks does what you asked: it reads, searches and downloads, and goes on through the steps your ask needs (a time slot, the cart, the checkout). The one step it holds is paying: I ask you the price once, at that step, and a look running while you're not there never pays.
3. Two permissions a task can have: “submit” lets it buy, post, send or sign up on your behalf, and it takes every step that needs — choosing options, the cart, the checkout and the final order — doing only what you asked; “offsite” lets it follow the goal onto other public websites (private and internal addresses stay blocked whatever you allow).
4. How far that goes today: “offsite” is enforced by the browser's network fence. Without “submit”, the browser lets each click, key press and form entry run except the one it recognizes as paying or placing an order — but it judges by what the page shows, so a button whose purpose is hidden (a plain “Continue” that places an order, or a page script) can get past it: that hold is a check, not a guarantee. With “submit”, your one yes covers the whole task, however many clicks it takes.
5. Allowed to act, it does the whole job: it picks what you asked for, fills in what you told it and goes through the checkout — and if you've connected your account there and it holds a card, it pays with that and places the order. It never types a password, a code or a card number itself — it refuses to type into those fields, whatever you allow (the one exception: if I ask you to text me a sign-in code, the one you send back is entered for you, once, and I don't keep it; passwords and card numbers are only ever typed by you) — so when the checkout needs a new card, you'd have to sign in or make an account, or a form asks for details you haven't given it, it gets everything else ready and hands you just that step in the secure remote browser (if that browser is switched off, it stops at that step and I tell you). The only personal details it types are ones you gave it for the task (a shipping address, say). It never signs in where you haven't given me a login. To have it work inside an account of yours, give me that login once — see “Connected sites”: I sign in with it by myself (it types your saved login into the sign-in fields; I never see it), and if you're signed out later, I sign in again the same way when you ask. The remote browser gets past a captcha by itself where it can; one it can't pass on a task that acts for you is handed to you as one quick step, and the task carries on once you've passed it (one running with nobody there tries another browser first, then I tell you; a task that only looks stops at it and I tell you) — unless it met one somewhere it only wandered to by itself, which it leaves to go another way, or, on a task that only looks, on one target while there is another one to read — one it already read, or one it was sent to that it hasn't reached yet, even when the check is on the first page it opened — which it reports as needing you and then finishes the rest. It's told never to solve one itself. On a job you asked for, though, a check nothing of mine can get past sends me to look for another site that does the same job first, and you get one question about it — I hand you just the check when you'd rather stay on that site, if no other site can do it, or if the work is in your own account there.
6. If a task may need you for a step only you can do — signing in, a code sent to you, a “verify you're human” check — I can start it so it pauses there instead of stopping: it runs in a secure remote browser, and you get a link to take that browser over — it waits for you up to an hour, as long as the task still has the time. The page says which site the browser is actually on and warns you when it isn't the task's own — only sign in where you expect to; I never ask for a password or a card number for myself. You do the step yourself — I don't touch the browser while you have it — then hand it back (tap Done there, or just tell me you're done) and the same task carries on in the same browser, from where you leave it, signed in if you signed in, following the original task’s navigation permissions. That page says so before you sign in. If the remote browser can't be opened, the task stops at that step instead of pausing, and I tell you why. A code emailed to you for that step, though, the browser finds in your Gmail or Outlook and types itself (only one sent from that same domain, and I never see it) — no take-over. If I ask you to text me a sign-in code, the one you send back is entered for you, once, and I don't keep it; unless I ask, a code you send me is just a message. When a step needs only your own details — an address, a phone number, or a traveller's date of birth, gender or title — I ask you in our chat instead of handing you the browser (if you asked for the task here, in Telegram, WhatsApp or iMessage, not by plain text): your reply is typed only into the page's matching fields, and our chat keeps a placeholder, not your words. Passwords and card numbers are only ever typed by you.
7. One yes covers the job — and before the click it recognizes as placing an order or paying, I show you the order as the page shows it (what it's buying, the total, the card — whatever it could read) and wait for your yes, unless you've set purchases to Buy when I ask and asked for exactly that order, you named what it buys and it's under the “don't ask under” amount you set there, or you said yes to the one-line offer I showed you for it — what, where and the total — and the order is still within that total, in the same place (a yes to an offer I only described in words still gets the order shown). If you left the choice to me (“get me a phone charger”), neither of those settings skips that: I show you the one I picked first. If an earlier step asked you first (saving an address, say), I ask again at the order, and if the order changes after your yes, I ask once more. It recognizes that click by what the page shows, so one a page hides can get past it. If the browser it runs in can't wait for your answer, it stops with the order ready for you instead of placing it.
8. If a website refuses the browser outright (an “Access Denied” page or a bot wall), it tries again on its own — first from the secure remote browser, then, if that's refused too, from a residential network — and you hear only the result; if every one is refused, I tell you.
9. Some stores and other places online put up their own notice that their terms don't allow AI agents or automated access, with a button to continue. I press their own button and carry on, the way you would — never another way around it. I do that every time the notice shows, however often it comes back. Only your own rule stops me there: to stop somewhere, say so (“stop shopping on Amazon”), or set “always ask me first” or “never” for it in Settings › How it works for you: there I stop at that notice and ask you instead, and only your yes (from the xhuman app, iMessage, Telegram or WhatsApp; a plain text from your number can't be taken as your yes) keeps going past it there again.
10. Want to see it? I can show you what the browser sees: ask, and it saves pictures of the page, or a part of it, and they come to you with its answer, here or on your phone (up to four in a message; one on a plain text message). It never keeps a picture of a sign-in or payment page, or of one where a password, a code or a card was typed.
11. If a purchase you approved fails before anything is bought (an error page, say), I try it again another way to the same order without asking you to start it again, and again if that fails too, as long as nothing was bought — it pays that order the way the first try would have, on the same yes and within the same limits, and asks you only where the first try would have asked. If it can't get through, I tell you.
12. Once started it runs on its own, usually for a few minutes. A running browser task has a Stop on its card, the × in its corner (in Telegram, a Stop task button); on a text, just tell me to call it off. While it's paused for you to take it over, that page has a Stop button too, which works from any of your devices. A Stop ends the run before its next step — a click already under way when you press it still happens.
13. Rarely, a task has to run on the older browser, which checks none of this — its read-only rule is only an instruction there. When that happens I tell you when I report the task.
14. Ask me “is it done?” and I check it: how far it got, the pages it went through, what it found, and download links for any files it saved.
15. You don't have to ask: when a task ends I tell you in the conversation you started it from, on my own — what it found, or why it stopped and what you need to do. If you started it in Telegram, Messages or WhatsApp, I answer right there in that chat (in your quiet hours it comes right away if you've written me since they began, silently on Telegram, and otherwise waits until they end; if you've told me to say nothing tonight, it waits until that ends, or until you write me), and only if that chat can't take it does it go to your next chat. If you started it by emailing me, where the answer comes back is under “Whether a result reached you”. If you started it here, the answer shows up in this conversation as soon as it's done. If you haven't seen it there within about 30 seconds (2 minutes if you were in the conversation in the 10 minutes before, say with this tab in the background), I send it once, to the chat you last wrote me from that can take it, not to all of them; only if that chat can't take it does it go to your next chat, one at a time, and only if no chat takes it to every browser you turned notifications on in. With no chat linked, it goes to those browsers, or your email when you have none of those — never while you've switched proactive notifications off. If you already asked me and I told you, I don't say it again. (On the web that holds only if you were in the conversation after my answer; if you'd left, your phone gets a short notice.) If none of those could take it, I still email it to you — when, and what happens if even that can't reach you, is under “Whether a result reached you”.
16. When a task finishes, my answer usually comes with a picture of the page it ended on: in Messages it's part of the same message, in Telegram and WhatsApp it's a file right after it (WhatsApp only within a day of your last message there), and here it shows under the answer in our conversation. Not for a sign-in, not with an answer that comes by email, not from a task that ran on the older browser, and not when a page tried to give me instructions. I keep the picture for 7 days.
17. Everything I do on my own, area by area — and what you agreed to — is in Settings › How it works for you; ask “what do you do without asking me?”.

Try asking

• Check their careers page and tell me which roles are open
• Open this link and summarize what the page says
• Check this site and tell me when their next event is
• Download the latest annual report PDF from acme.com/investors
• Is the browser task done yet?

Checking several sites at once

How to

1. Just ask ("find me a quiet Italian in the West Village for two at 7:30, about $100 a person"), and I look at the booking sites for the best places at the same time.
2. You get one message, never one per site: the places that fit, each with what it has open and its price next to your budget, and my pick.
3. To book one, say which, or just say yes to my pick.
4. A site that turns the browser away doesn't hold up the others: I tell you what the rest found and that one site couldn't be checked.

Try asking

• Can you check a few booking sites at once?
• Compare tables on OpenTable and Resy at the same time
• Look at several sites for a hotel room and tell me the best

Redirecting a browser task

How to

1. Tell me in your own words while it runs: “use the cheaper option instead”, “skip that site”, “look for the blue one”.
2. It picks the change up before its next step, usually within a few seconds.
3. To stop it, just tell me to stop.
4. A task you already said yes to (a booking, a purchase) takes your change the same way, on that same yes. Once it has already sent or posted something for you, a further change goes on only in words you type yourself, in the app or on iMessage, Telegram or WhatsApp (a plain text from a number that can't prove it's you can't carry it to new people).
5. A task that already finished can't be changed. If it stopped short in the last couple of hours without placing anything, I start that same job again with your change, on your same yes; otherwise the change is a new task.
6. You can change a task up to five times while it runs.

Try asking

• Use the cheaper option instead
• Skip that site and try another one
• Look for the blue one
• Change the browser task to only check prices under $50
• Actually, tell it to stop at the first result

Watching a browser task live

How to

1. In the web chat, the task's card opens on the live page by itself, with my moving orb until the first page shows. Tap the page for full screen, minimize it with the round button, or stop it with Stop. There's no list of steps; the one line under the chat says what I'm doing.
2. If you scroll away while I need you, an orange badge appears by the message box; tap it to jump back to what I need.
3. On your phone I don't send a link when I start: you hear from me when it's done, or when I need you. If you'd like to watch, just ask and I'll text you the link to the same live view.
4. Only you can watch your tasks, signed in. The cursor (a dot on the watch page) is where I last clicked or typed.
5. You see the whole page live, all the time. I cover password, card and code boxes with dark blocks, and also boxes I filled from your saved logins, a payment company's card panel as a whole, and, after you've taken the browser over, the boxes you typed in. If part of the page is slow to load from another site, it can show dark for a moment. When the page is changing or I'm typing something secret, the last picture just stays up until the next one.
6. Watching never steers or pauses it. To do a step yourself, use Take over; to change course, just tell me.
7. When I book or buy something, I keep a picture of what the site showed right after, usually its confirmation, and send it with my message. It stays in the task's card in the web chat too. Password, card and code boxes are covered on it.

Try asking

• Can I watch the browser while you work?
• Show me what you're clicking
• Where's the live view of the browser task?
• Can I see the browser live?

Interrupt me

How to

1. Just type it. On the web, send the message while I'm working; on your phone, text it. You don't have to wait for me to finish or press stop first.
2. "Stop" works: if I haven't sent or done the thing yet, I won't. If it has already gone out, I say so plainly rather than pretending it didn't.
3. Additions count too: "cc Dana", "make it Thursday instead", "leave out the pricing". I fold them into the work in progress instead of starting over.
4. Before anything I'd do that can't be taken back (sending or deciding something in your name, including acting on your okay of a waiting draft, paying, posting, deleting, joining a call), I read any new message you sent first. Reading and drafting carry on meanwhile. That holds for a photo or file you send too. A plain yes or 👍 straight on a draft I texted you sends it right away, so a change of mind after that comes too late for that draft. If you message me during a scheduled or background run, your own turn answers right after that run; its suggestions still reach you as cards. One long single step (a browser task) reads your message when it returns. "Stop everything" still stops at once.

Try asking

• Can I change my mind while you're still working on it?
• What happens if I tell you to stop halfway through sending an email?
• If I add someone to cc while you're drafting, will you include them?

Booking another way when a site blocks me

How to

1. Just ask me to book it — you don't need to name a site.
2. If you name a site ("on OpenTable"), I stick to it; if it blocks me, I offer the other way in rather than taking it.
3. I only switch the way in, never the place, the time, the party or the price you agreed to.
4. If every site blocks me, I call the place — only while it's open, and I never pay or give a card on the call.
5. You get one message with what I booked and where, plus how to cancel it.

Try asking

• If OpenTable blocks you, can you book it another way?
• What happens if the booking site won't let you in?
• Can you use Google to reserve the table instead?
• If the site blocks you, will you call the restaurant instead?

Buying at another store when one blocks me

How to

1. Just ask me to buy it — if the store turns me away, I look elsewhere on my own.
2. You get one message: the item, the other store and its all-in total, and the link to finish at the first store yourself.
3. Say yes and I check out there; if the total at checkout is higher than I said, I ask you once first.
4. If no other store has the same item, I tell you so and give you the link to finish it yourself.

Try asking

• If the store won't let you check out, can you buy it at another store?
• What happens if Amazon blocks you from buying it?
• Can you find the same item at a different store?

Remembered sites

How to

1. It's automatic: each site I browse for you in my secure remote browser remembers you alone, site by site, and keeps only that site's own data.
2. See them in Settings › Vault › Remembered sites, no unlock needed: each site, since when, when I last used it, and whether it may keep you signed in.
3. Tap Forget on a site to clear what I kept for it — a later visit starts a new one. If a site turns my browser away, I clear what I kept for it myself so the next visit starts fresh (unless it may be keeping you signed in). A site I haven't used in 90 days is forgotten on its own, and deleting your account clears them all.
4. If you sign in on the site during a task, or I sign in there for you (say, with a code it emails you), that sign-in stays in the site's memory, the way your own browser keeps it, until you Forget the site — and from then on I treat my visits there as signed in to you. Saving a sign-in on purpose is Connected sites.

Try asking

• Do you keep cookies when you browse for me?
• Which sites does your browser remember?
• Clear the cookies you kept for opentable.com
• Why does the site remember me now?

Research jobs

How to

1. Ask for it in words (“research the top five competitors to Acme and come back with a comparison”) and it starts right away, running on its own for up to two hours.
2. A research job reads the place you ask for: the open web, or your own records and connected accounts — even after the conversation has read something from outside (an email someone sent you, a web page). For work that needs both, I start a helper instead, which works with your whole toolbox.
3. You don't have to wait or ask: when it ends, I tell you in the chat you asked from — what it found, or why it stopped (out of budget, out of time, couldn't finish). Its result is checked by code against what it actually read before I call it done, and anything it quotes from the web is shown to me as outside content, never as an instruction.
4. Ask “how is my research job going?” and I check it: whether it is still running and how long it has left. Say “stop it” and it ends at once — a stop is silent. You can also tell it something new while it runs (“skip paywalled sites”), and I pass it on at once.
5. Stop everything also stops every running job.

Try asking

• Research the top five competitors to Acme and come back with a comparison
• Can you run a long research job in the background?
• How is my research job going?

Background helpers

How to

1. Ask in words (“while I'm in meetings, sort out my travel receipts and draft the expense report”) and a helper starts right away — no card to approve first.
2. Give me a list (“for each of these five vendors, find their cancellation terms and draft the cancellation email”) and I start one helper per item, however long the list, all working at the same time. Each one tells you here when it ends.
3. A helper does what you asked with your own tools, sends and deletes included. If it needs to pay for something, it doesn't — it leaves that waiting for you as one card with the price, and says so when it reports back.
4. It checks each thing it does against what YOU asked for when you started it. Something that doesn't match — say, a web page it read told it to email someone else — is skipped, and it tells you.
5. Ask “how are the helpers doing?” and I check them; say “stop it” and one ends at once; you can tell one something new while it works (“use my work email”). Stop everything stops them too.
6. A helper can't ask you questions mid-way or start more helpers, and it browses websites like any work that runs on its own: looking always, acting when your words asked for it on that site, never on a site you turned Off (Settings › How it works for you › Websites).

Try asking

• Can you work on this in the background and tell me when it's done?
• Put a helper on each of these five vendors
• Run these tasks in parallel with background helpers
• How are my background helpers doing?

Wide research

How to

1. Ask in words (“for each of these 12 companies, find the CEO and the latest funding round”) and it starts right away.
2. Any number of items. A repeated item (same words, ignoring case) runs once, and the report says how many I dropped. Past 20, I split the list into even runs of at most 20, and each run brings back its own report.
3. Workers start a few at a time, a couple of seconds apart, and each answers in the same fixed shape (the fields you asked for, plus its sources). An answer that doesn't fit gets one retry; a worker that fails gets one retry; after that the item is reported as not done, with the reason.
4. You hear back in the chat you asked from when every item in a run has settled, once per run: the totals first, then what each found. Answers that couldn't be fully confirmed against what the workers read are marked as first passes. Stop a run and every worker in it stops at once; Stop everything stops every run.
5. It's the bigger sibling of a research job, so it's available wherever research jobs are. A conversation that has already read outside content (an email, a web page) can only start a web run.

Try asking

• Research each of these 12 companies and find the CEO and the latest funding round
• Research each of these 8 people in parallel
• Can you run the same research on a whole list at once?

Connected sites

How to

1. Just ask me in any chat — “log in to my Uber”. If I don't have a login for that site yet, I use an email or phone I already have on record for you (a mailbox you connected, the number you text me from, your account email) — no question first; only when I have none do I ask which email or phone your account is under. Then I sign in with it myself in a secure remote browser, with no link to tap. If you'd rather not say, or the site needs more than that, I send you a link where you type your login once (the page is built not to look like an xhuman sign-in, so your phone shouldn't offer your xhuman password there): it goes straight into your encrypted vault, never through chat — I never see a password. If you already saved a login, I sign in right away. You can also add one yourself in Settings › Vault › Add a login, which doesn't need the vault unlocked.
2. If the site emails a code to a mailbox you connected, I read it and finish by myself. If it shows a “verify you're human” check the browser can't get past by itself, or texts you a code, I send you a link to that one step on a screen you can tap and type on from your phone; tap “Done” and I finish signing in. If it asks for a password after all, the link asks for just that password — it's added to your saved login, and I type it. I never ask for a password or a code in chat. If you'd rather sign in yourself (say, with Google or Apple), the same page lets you, and you press “I'm signed in”.
3. Ask for something there in the same breath — “log in to my Hacker News and tell me my karma” — and it runs the moment you're signed in, signed in and only looking; its answer is the one message you get, no second ask. An order, booking or post you asked for the same way (“order me a burrito bowl on Uber Eats”) carries through the sign-in too: once you're in I tell you I'm starting it, and before I pay for anything I ask you, with the total when the site shows one (under a “don't ask under” amount you set for that site, something you named goes through without asking, but when I'm the one choosing which, I check with you once first). You never have to type it again.
4. If a site has moved to a new address (notion.so is notion.com now), I sign you in where it lives today.
5. Your sign-in (the site's cookies) stays in that private browser profile until you disconnect the site. Disconnect deletes it at Browserbase; your saved login stays unless you forget it too (Disconnect offers that). If the site signs you out later, just ask me to log in again — with your saved login I sign back in myself, no link to tap; you only hear from me if it wants a code or a quick check the browser can't get past by itself.
6. After that, a browser task whose start page is on that exact site runs signed in as you, and says so: “Signed in to uber.com as you”.
7. When you ask me to DO something there — “order me an Uber to JFK”, “buy this on Amazon”, “book the table” — I act for you: I do the steps myself (where to, which option, the cart). I ask you the price once, at the step that pays, naming the amount and where it goes, unless you already said yes to that price in our conversation; a send, post or delete you asked for goes ahead with no second question. One yes — a tap, a 👍 or “yes” in your chat — and it's done; I tell you the site's confirmation. When you just want me to look something up there, the task only reads.
8. Your guardrails, per site in Settings › Vault (open the site): never act on this site (tasks there only read), or don't ask under an amount you pick. A task never goes to other sites while signed in, never types a password, and runs only when you ask in a conversation — never from an automation — one task per site at a time.
9. Some sites can't be connected: email and calendar providers, sign-in services (Google, Microsoft, Apple…), and xhuman itself. A site you sign in to THROUGH one of those — with “Continue with Google” or “Sign in with Apple”, Microsoft or Yahoo — can still be connected when you ask me to sign you in there: I open the site, and on the provider's own page you finish signing in yourself, approving it on your phone if it asks; I never type there. The site stays signed in, and your Google or Apple sign-in itself isn't kept in that browser — only the site's.
10. Pause everything, or a task's Stop, stops a sign-in too.

Try asking

• Login to my Notion
• Log into my supplier portal for you
• Which sites are you signed in to?
• Forget my amazon login
• Do you store my password?
• Can you buy on amazon signed in?
• Order me an Uber to the airport
• Check my Hacker News karma
• Log in to my Hacker News and tell me my karma
• Log into my X and tell me what matters in my For You feed
• Log me in to Notion with my Google account

Ordering, buying and booking for you

How to

1. Ask in a conversation, in your words: “order me an Uber to JFK”, “buy this on Amazon for me”, “book the table for 4 at 8”. If the site isn't connected yet, I sign you in first — with an email or phone I already have on record for you, no question first (only when I have none do I ask which email or phone your account is under), or with a secure sign-in link when a step needs you — and your order picks up by itself once you're signed in — you don't ask twice. You can also sign in to a site ahead of time in Settings › Vault › Add a login.
2. I do every step of what you asked myself — where to, which option, the cart, the checkout page, a card saved on your account there, if it holds one, and a post, a send or a delete you asked for. I ask you once at the click I recognize as placing the order or paying, naming the amount, the card and where it goes as the site shows them, and once more if the order changes after your yes. A button that may place the order but doesn't say so, I ask you about once, by name. One yes — a tap, a 👍 or “yes” — and it's done; then I send you the site's confirmation (order or trip number, arrival time).
3. If the price goes up or the destination changes after your yes, I don't press it — I ask you again with what changed. Something on a page asking me to buy or send something else is refused: every such step is checked against what you actually asked for.
4. Guardrails, per site in Settings › Vault: never act on this site, or don't ask me under an amount you pick. How purchases are confirmed (a reply, or only a tap) is in Settings › How it works for you › Money.
5. I never type your passwords or card numbers: a login you saved is typed into the page by code, and I never see it; a card number you type yourself. Email and calendar providers and sign-in services (Google, Microsoft, Apple…) can't be connected, so I never act signed in there. And I act signed in as you only when you ask in a conversation, never from an automation. A site you sign in to with Google or Apple (Uber via “Sign in with Google”) can be connected that way: you finish Google's or Apple's own sign-in yourself, and I keep only the site's sign-in, never your Google or Apple one. If I ask you to text me a sign-in code, the one you send back is entered for you, once, and I don't keep it.

Try asking

• Order me an Uber to JFK
• Buy this on Amazon for me
• Book the table at 8 for four

Saved logins

How to

1. Ask me to log in to a site (“log into my X”): I ask which email or phone the account is under (that much you can tell me in chat, and I save it), or send you a link where you add the login — or save one in Settings › Vault › Logins › Save a login, leaving the password empty for a site that only sends codes. Never paste a password in chat; if one lands there anyway, I'll send you the secure link instead.
2. It's stored encrypted under your own key. I sign in by going through the site's own sign-in page, and the browser types the saved username and password into its fields itself — only on that site, and only into its sign-in form (never a “confirm your password” form for changing settings). The model (me) never sees them.
3. If something I'm doing for you runs into a site's sign-in page and I don't have a login for it, I don't hand you the browser: I send you one link to save that site's login once, then I sign in and carry on with what you asked — an order still asks you before it pays. With a login already saved, I just sign in and keep going. (On an account that only signs in to xhuman by text, that step comes as a screen to sign in on instead.)
4. If the site then wants a code it emails you, the browser finds it in your mail and types it (only a code from that same site). A code texted to you, or a “prove you're human” check, comes to your phone as a link to one screen you can tap and type on — then tap “Done”. If I ask you to text me a sign-in code, the one you send back is entered for you, once, and I don't keep it. If the site rejects the password, you can save it again on that page. If a site wants a password and I only have your email or phone for it, that link asks for the password alone and adds it to the saved login. Codes are never saved.
5. Which logins are saved: they're listed in Settings › Vault › Logins. I can't read them, so I can't tell you what's in there.
6. To change one, open its row in Settings › Vault and choose Change password or Replace login: what you type is saved without being shown back. To forget one, just tell me (“forget my Target login”) and I delete it at once — or open its row in Settings › Vault and choose Remove. It can't be brought back, and they all go if you delete your xhuman profile.

Try asking

• Save my zoom login
• Which logins have you saved?
• Forget my zoom password

Your vault

How to

1. It's in Settings › Vault. I can't read what's in it, so I can't list your passwords or card numbers for you.
2. Logins: one row per site, saying whether you saved the login or I made the account (and with which email), whether I'm signed in there, and when I last used it. Open a row to change its password, replace the whole login, sign me out of that site, or remove it. Removing asks you once, can't be undone, and signs me out there too. Saved job-site logins are listed there as well, each with Remove.
3. The Vault is locked until you unlock it with two quick checks: signing in again (your xhuman password, Google, or a code to your email) and a code texted to your phone. It stays open for 10 minutes, and locks again when you leave the page, hide the tab for a minute, press Lock or sign out. Once it's open you can see and copy the username and password of every login there, the ones you saved and the accounts I made (with any earlier password one of them had). A new password you type there is saved without being shown. Every login also has "Forward its emails to you" for what the site sends to my address, the ones you saved as well as the accounts I made; it stays as you set it when you change or replace the login.
4. Cards and payment: your Link wallet (when it's connected I pay with Link first, and you approve each purchase in Link), the one card saved in your vault (its brand, last four digits, expiry, the name on it and its billing address, never the full number, with Replace and Remove), and the card your plan and credits are billed to (Change opens Stripe's page to update it).
5. Keys: an API key you gave me for an app I connected to can be replaced with a new one (it's never shown back) or removed, and it says how many AI apps or API tokens can use your xhuman, which you disconnect in Settings › Apps (the Chrome extension on its Your Chrome row, the rest under Old access).
6. Passports, IDs and travelers: the same page has Passports & IDs and Travelers, for bookings that need them (flights, visas, hotels). "Add a passport or ID" takes a photo or PDF of the photo page (the coded lines are checked digit by digit) or typed details; "Add a traveler" takes a person's legal name, birth date, gender as on their ID and, if you like, a passport or ID, Known Traveler and frequent-flyer numbers, for you, your child or anyone you book for. Adding works on the locked page and shows nothing back; once unlocked each item shows only a masked view (a passport's last two characters and expiry month, never the full number), with Replace or Edit and Delete, one tap, gone for good. They're sealed under your own key and kept until you delete them, and a booking uses them without asking you again. You can also send me a photo of your passport in chat, and "delete my passport" in chat removes a saved one. To delete a traveler, open Settings › Vault and tap Delete on their row.
7. Details for forms: the name, email and phone I fill in on a checkout or booking you asked for. Change them in Settings › General.
8. Add a login signs xhuman in to a new site right here, and it doesn't need the vault unlocked; each site also has Sign in again, Sign xhuman out, and what xhuman may do on it.

Try asking

• Where can I see everything you've saved for me?
• Open my vault
• Where do I add my passport?
• Delete my son's traveler profile
• Where do you keep my passwords and cards?

Paying with your saved card

How to

1. Ask me to add your card and I open the secure card form right here in the chat (on a phone, one link that brings you straight back to this thread) — you type the number, expiry, the name on the card and your billing ZIP there, never in a message. Afterwards it shows only as its brand, last four digits and expiry. You can keep several: the form lists your cards with “Add a new card”, “Replace” on each, and “Make default” on the others — the Vault page has the same.
2. Of the cards in your vault, a purchase pays with your default unless you name another — say “use my Amex” or “the card ending 1005” and that card pays; the question with the total names exactly which card (“…on your saved Visa ••4242”). A card you just added becomes your default.
3. To change your default, just say so in the chat (“use my other card from now on”, “make the Amex my default”): I make that card your default right away, with no form, and every purchase after pays with it unless you name another.
4. The security code is never saved. If a checkout asks for it, after your yes I send you one secure field for it — typed there for that one payment, entered by the browser, then gone.
5. If a purchase reaches a checkout before you've added one, the link I send you opens on that same add-card form, naming the order it's for; save it there and I carry on with the order.
6. In the web chat, the card form opens right over the chat — the same encrypted fields as the Vault, with the order and its total on top — so you never leave the conversation. Where that purchase can take it, you can type the card's security code there too: it goes to that one purchase and is never saved, so I finish without stopping to ask for it.
7. No card saved yet and Pay with Link is on for you: the message with the total asks “Link, or add a card?”. Say add a card and I send you the add-card form's link; once it's saved I pay that total with it on the yes you already gave — no second question at that total.
8. Every purchase still asks you first, naming the item, the store, the total and “your saved Visa ••4242” — that question is never answered by a setting, only by your own yes: a tap, a 👍, or a yes typed in iMessage, Telegram or WhatsApp, never a code texted from a plain phone line. If it costs more or anything changes — while you type the security code included — I ask again or stop, and nothing goes through without a new yes from you; a card already saved on your account at the store is used first.
9. The card is filled only into that store's own checkout page — including the store's checkout on a payment company's page, when the store itself sent the browser there — or its payment provider's card box, whenever that checkout asks for it after your yes (there's no time limit while the purchase is still going), and emptied from the page right after.

Try asking

• Add my credit card
• Add another card
• Pay with my Amex instead
• Use my other card from now on
• Can you pay with my card?
• Where do I put my card?

Filling in an address

How to

1. Start typing the street: after three characters, up to five matching addresses show under the field. Tap one, or use the arrow keys and Enter, and the rest of the address fills in; you land on the apartment field in case you want to add a unit.
2. Your saved home address and your saved card's billing address show first: on a phone as soon as you tap the empty field, on a computer once what you type matches one (or press the down arrow).
3. You never have to pick one: whatever you type is kept as you typed it, and an address that doesn't show up is typed by hand as always. Your browser's own autofill and your password manager still fill the same field.

Try asking

• Can it fill in my address for me?
• Why do addresses pop up when I type my street?
• How do I fill in my billing address faster?

Paying with Link

How to

1. You don't have to set it up first: pick Link when I ask how to pay, and I send you Link's sign-in in the same chat — sign in and the order carries on by itself. You can also connect it ahead of time in Settings › Vault › Cards and payment, and disconnect it there anytime.
2. Every purchase paid with Link is approved by you in Link, never by a setting or a yes in this chat: I send the order, its total and Link's approval link together.
3. The one-time card is for that store and that total only. If the total moves past what you approved, I ask again; if the task ends before it's used, I cancel the request in Link.
4. The message with the total names how I'll pay: Link once it's connected, otherwise the card saved in your vault. With nothing connected it asks “Link, or add a card?” and pays the way you answer; a card you add there is paid on that same yes once it's saved. A card your account at the store already shows is used first.
5. Once Link is connected I pay with it by default, so there's no security code to type. Say “with my card” (or “with Link”) when you ask and I pay that way instead.

Try asking

• Can you pay with Link?
• How do I connect my Link wallet?
• How does paying with Link work?

Signing up for you

How to

1. It happens on a browser task you've let act for you (buy, book, sign up): if the site asks for a new account to get it done, the browser fills in the sign-up itself.
2. It uses my address (your handle @agent.xhuman.com) unless you'd rather it used your own email — say so once and it's remembered for every later sign-up.
3. The password is generated by code, saved to your encrypted vault first, then typed by the browser — never shown to me or written in chat. The account is listed in Settings › Vault, and the browser can sign in with it again later.
4. The account is yours: unlock Settings › Vault (two quick checks: your password, Google or a code to your email, then a code texted to your phone), open the account and you can see and copy its username and password, so you can use it in the site's own app; if I ever had to redo a sign-up, the earlier password is there too. Settings › How it works for you › Websites › Email for new accounts picks which email I sign up with.
5. If you already saved a login for that site, it never makes a second account or replaces yours — it hands you the sign-in instead.
6. A confirmation code the site emails to YOUR address is found and typed by the browser. One sent to MY address during the sign-up — a code, or a "confirm your email" link — is used by the browser for that sign-up only: only mail from that site, checked by the mail server, while that task runs; it never opens an unsubscribe, payment or password-reset link, and nothing in that mail is ever taken as an instruction. If nothing arrives in time, that step comes back to you. A phone number or anything else you didn't give stops and asks you.
7. Everything that site sends to my address afterwards — your e-ticket, the receipt, a password reset — is forwarded to your own email exactly as it arrived (attached), after the mail server checks it really came from that site. I never act on it. To stop it for one site, turn off "Forward its emails to you" on that account in Settings › Vault.

Try asking

• Sign up for the Acme newsletter for me
• Make an account on that site and book it
• Use my own email when you sign up for things

Codes a site emails you

How to

1. Nothing to set up: it works on any browser task — including when I sign you in to a site with your saved login — with the mailboxes you connected (Gmail or Outlook). Your request covers it — I don't ask again, and I don't ask you to send me the code.
2. Only an email FROM the site the browser is on, received since the task started, is read, and the code is typed only on that site's page — a code one site sent is never typed into another.
3. The code never reaches me, the task's history or the logs: the browser types it and goes on.
4. If the site turns the code down or it has expired, the browser presses the site's own “send a new code” button, where the page has one, and reads your mail again for the new code — up to twice — and only then hands the browser to you for that step. If no such email has come yet on a task you can watch live, I ask you for that step right away and keep reading your mail, so a code that arrives is still typed for you; on a task you can't watch, the browser asks the site for a new code the same way first. A code from a different site is never used. Codes sent by text message aren't read from your phone. If I ask you to text me a sign-in code, the one you send back is entered for you, once, and I don't keep it.

Try asking

• Log in to my account on that site and check my balance — it'll email me a code
• Can the browser enter the code the site emails me?

Codes you text me

How to

1. Nothing to set up, and nothing to do unless I ask. I ask only for a sign-in you asked me for in our chat, with a login you saved — or for a code step in the middle of something you asked me to do in our chat, like a booking's email verification — when the site's own page is showing a code box I can recognize: you get one message asking for the code that says in one line why I'm asking (where the site sent it, and that it isn't in the inboxes I can read). If I can read the code from your email, I enter it and don't ask. Partway through a task I ask only when the page is plainly a booking's, a checkout's or an order's code step; a code to change your email, phone, password or payment details, a security step, or a code that links a TV or another device always comes to you as a link so you see the page yourself. Otherwise the code step comes to you as a link, as before.
2. When I've asked, send just the code on its own (482913, or 482 913) from any chat you've linked, or here on the web: the browser task's card shows “Enter your … code” — tap it, type the code in its one box and press Submit, and it says “Code submitted ✓”. It is entered into that site's code box once. I never see it and I don't keep it: our chat shows “(code received)” in its place. (Your phone company or messaging app keeps its own copy of what you send, as with any message.)
3. A message that isn't just a code — a question, or the code wrapped in a sentence — is read as an ordinary message and nothing is typed; send the code alone.
4. If the site doesn't accept it I tell you once and you can send a new one, one more time. The ask stays open as long as the site's code step waits for you, up to about 60 minutes, and ends as soon as the sign-in finishes or moves on; after that a code you send is just a message.
5. Only one-time codes a site sends you. I never ask you for a password or a card number, and those are only ever typed by you, on the site's own page.

Try asking

• Log in to my account there — it'll text me a code
• Can I just text you the code the site sends me?
• The booking site emailed me a verification code — can you enter it?

Background jobs

How to

1. Every such job ends in one closing: it finished, it couldn't finish, or — if a deep search or a job scan stops reporting and I can't tell whether it finished — once, that I lost track of it, so you can ask me to run it again. I never assume it worked.
2. Where you hear it: if you asked in Telegram, Messages, SMS or WhatsApp, the answer comes back in that chat, and only if that chat can't take it does it go to your next chat. If you asked here on the web, its card in the conversation fills in and is the answer — a browser task also tells you in the conversation. If you asked by emailing me, where the answer comes back is under “Whether a result reached you”.
3. If a job's closing couldn't go out on time (an outage on our side) and it's hours late, I still tell you when it's done: in the app the message says first that it's late and why, then what the job found or did; on your phone the result comes first and a short late note follows it.
4. Ask “what are you working on?” and I list what's live for you, each from its own record: the deep searches, job scans and browser tasks still running, your standing rules and watches, my check-backs and anything scheduled to go out, each said as it really stands — and any waiting on you, like a deep search that needs you to say which person you meant. A business call or a task on your computer isn't in that list; your Running page shows those.

Try asking

• Will you tell me when the deep search is done?
• Did anything you were doing for me get lost?
• What are you working on for me right now?
• I asked for a job scan on Telegram — will I hear back there?

Whether a result reached you

How to

1. If every channel I tried refused the message — say you blocked the Telegram bot, or turned texts off — or failed to take it, I email it to you once, about five minutes later, saying which channel didn't take it. If you open the conversation first, there's no email. Never while I can't tell whether a message went through, never if you've switched notifications off, and never a second text.
2. If even that email can't reach you, I record that it was NOT delivered, and I tell you first the next time you talk to me, wherever that is.
3. A channel that refused on your own say-so — you blocked the bot, turned texts off — I stop trying until you message me there again (tapping Restart in Telegram counts) or relink it in Settings › Reach. Meanwhile results come on your other channels, or by email if there are none — never while you've switched notifications off. When a task's result comes by email because of it, the email says which channel didn't take it, and the first one says how to turn it back on.
4. If you asked by emailing me — a browser task, a deep search to research someone, or a job scan — I reply in that same email thread when it's done, to your own address only (never to anyone else on the thread), and the same answer comes to your other channels too. If that address is no longer one of yours, or your agent's inbox is off, I answer in the app and on your other channels instead.
5. That reply in the thread counts as delivered once it goes out; if your mailbox refuses it, I don't send a second one saying the same — it's recorded as not delivered, and I tell you first next time.
6. A message that waits for your quiet hours to end counts once it actually goes out in the morning roundup; if that roundup can't send it, it's recorded as not delivered too.
7. If you opened the conversation where the result is, that counts as you having it — I don't repeat it.
8. Ask “did anything you were doing for me not reach me?” and I check.

Try asking

• What happens if my Telegram is blocked when a task finishes?
• Did the result of my browser task actually reach me?
• Did anything you were doing for me not reach me?
• I never got the message about my job scan — did it go out?
• If I ask you by email to research someone, will you email me back when it's done?
• If I start a browser task by emailing you, will the result come back in that thread?

Proactive proposals

How to

1. Ask “what's waiting for me?” and I list the proposals the engine staged.
2. Text “drafts” and the emails I drafted for you arrive in the chat, each whole — 👍 sends exactly what it shows.
3. Answer in plain words — “do the first one”, “skip the reconnect” — and I execute or dismiss it; nothing runs until you say so.
4. Ask “why did you suggest her?” and I answer from the card's own reasoning — the goal and the path it was written from.
5. Tell me why — “wrong person, she left”, “not till after the 20th”, “stop suggesting intros” — and I stop suggesting them, until then or for good.
6. Ask “what do you think I'm chasing?” and I show the goals the engine works from — correct or retire any of them in plain words and it sticks.
7. A proposal can rest on something you said you can offer, or on what is going on for you — ask “why did you suggest I help Marc?” and I answer from the evidence on that card (the fact and the date it was true).
8. I keep track of every reply you owe and every person you're waiting on, and judge each day whether one is worth raising now, needs its draft refreshed first, or can wait. Ask “why are you reminding me about Jordan again?” and I answer from that conversation's own judgment (get_task's followUp: why it matters, how many times I've nudged, when I'll look again).
9. Say “stop reminding me about this” (or “let it go”) and I dismiss that conversation's task — your decision closes it and I stop following up; I never close one on my own.
10. When you still owe someone important a reply I nudge you again until you act — but never about the same conversation twice within 20 hours, and at most 3 of these reminders a day; ask “why didn't you remind me again today?” and that is usually why.
11. Everything I do on my own, area by area — and what you agreed to — is in Settings › How it works for you; ask “what do you do without asking me?”.

Try asking

• What did you find for me?
• Show my pending proposals
• Approve the outreach one, skip the rest
• That M&A campaign isn't my goal — drop it
• Why did you suggest I help Marc?
• Why did you suggest Dana?
• Not her — she left. And stop suggesting intros
• Why are you reminding me about Jordan again?
• Stop reminding me about this

Permissions

How to

1. Ask “what can you do without asking me?” or open Settings › How it works for you: it says what is handled on its own today, what always asks you first, and your own “always ask me first” rules, which you can edit, pause or remove there. I answer from your settings as they are, area by area, including the disabled ones — never from this list.
2. What waits for you is exactly what the page says.
3. Turn anything down there or tell me — “make my automations ask me first”, “turn automations off” — and it applies at once, on any turn. Your email replies are set on the same page and each campaign on the campaign (a campaign has no Off — pausing it is its own control).
4. Automations work “as today” until you choose: anything that would show you a card waits for you, while some changes happen on their own — the page lists which. Once you choose for all automations or for one of them, they follow it: Off — they change nothing on their own and only read and report; Approve — anything that reaches people, spends or can't be undone waits for you under “Waiting for you”, with the reason on the card. Stop everything holds them the same way while it's on. An automation you set to send without asking still sends its emails — that's set on the automation itself.
5. “Always ask me first before using amazon.com” or about one automation is a rule here, not a standing rule; about a person or a company domain it's your inbox's reply rule. Each rule says whether it binds today.
6. Letting it do more on its own — automations back to how they worked before — happens on your yes — saying so in chat is the yes, with no card: in the app, or typed in any chat, a plain text included (how a purchase is confirmed loosens only from the app, iMessage, Telegram or WhatsApp). Letting browsing act on its own is set for one site, one automation or every site, and applies at once; letting automations act on their own is its own step (see “Automations on Just run it”).
7. “What did I agree to?” lists every time you let it do more, in the words you read, and when.
8. A busy day never sends anything back to asking: there is no daily limit. If an automation waits for you, the page says why — your own setting, a rule, or Stop everything.
9. Everything automatic is lowered here and stays lowered; to stop everything that's running at once, that's Stop everything (“Pause everything” on the same page).
10. Debug capture (disabled by default; its switch shows on Settings › How it works for you to the xhuman team, and to anyone who has it enabled so they can disable it): when enabled, I keep the full request of each of my model calls — instructions, the conversation as sent, tool names, model settings — for 7 days, encrypted with your key, so a wrong answer can be re-run exactly. Disabled, only an id-and-size record (no message text) is kept, for 30 days.
11. Purchases (Settings › How it works for you › Money): before a remote browser task pays, it names the item, the store, the total and the card in your chat and waits — for one yes (the default: a reply, a 👍 in iMessage, a tap in Telegram or WhatsApp, YES 1 by plain text), for a tap only, or, if you choose, not at all when you asked for it yourself (anything I decide on my own, or from a page, an email or an automation, still waits). Loosening it is a tap on the words first; each yes covers one purchase for 15 minutes, and a higher total, another store or card asks again.

Try asking

• What can you do without asking me?
• Make my automations ask me first
• Always ask me before buying on amazon
• What did I agree to?
• Why did you ask me before booking?
• How do I enable debug capture?

Ask me before every send and every purchase

How to

1. Tell me “never send or spend without asking” (or “ask me before every send and every purchase”). It applies at once, on any turn, and I tell you in one sentence what now asks.
2. With it on I ask before every email, reply, invite or message — even one you tell me exactly, word for word — and before every purchase, even one under your limits or one you already said yes to in chat.
3. Anything an automation, or one of your inbox rules, would have sent on its own is drafted and waits for you instead.
4. Three things I normally send on my own are not sent at all while it is on: meeting prep to your guests, a running-late note to a guest, and scheduling replies on a thread you Cc me on (that thread is handed back to you, with the reason).
5. It does not pause what you already launched to run on its own: a campaign's sequence, the intro asks you approved for it and its autopilot outreach, auto-apply, a message you scheduled, booking-page confirmations and reminders, and credit auto-reload keep going until you pause them where you started them.
6. It does not touch reading, searching or drafting, a calendar event with no guests, or credits spent on something you asked for.
7. It's under Always asks you first in Settings › How it works for you (/settings/how-it-works#ask), as its own switch, and I can tell you whether it is on.
8. Only you can switch it off: tell me you no longer want me to ask before every send and every purchase, or use the switch in Settings › How it works for you. In a conversation that has read an email, a page or a file, switching it off waits for one tap from you on your approvals list — a page, an email or an automation can never switch it off.

Try asking

• Never send or spend without asking
• Ask me before every send and every purchase
• Stop asking me before every purchase

Things I handle on my own

How to

1. Approve the same kind of email as written a few times and I ask you once: want me to handle these on my own? Say yes and it becomes a rule.
2. Or tell me in chat, for example “handle scheduling emails for me” or “stop asking me about confirmations”. I save it as a rule in plain words and it's on from then.
3. Every rule is listed under What it does on its own in Settings › How it works for you (/settings/how-it-works#own), on the row for what it covers.
4. Edit a rule, pause it or remove it there, or ask me to. A paused rule asks again until you turn it back on.
5. Money, buying and deleting for good always ask, whatever rules you have.

Try asking

• Can you just handle scheduling emails for me?
• Stop asking me about confirmations
• What rules do you follow on your own?

Open to

How to

1. To join xhuman, just text me from your mobile, on iMessage, a regular text or WhatsApp: I ask for your name and email right in the chat and make your account there, no website needed. From a regular text I first send you a six-digit code; text it back to prove the number is yours. You can also start at xhuman.com/start with your name and Google or your email, then text me once to link your number. A number already on another account can't be linked to a second one.
2. My first message lists what I can take on for you, numbered: your inbox first (with your email connected I can see what matters most and make the rest personal), then a morning brief, finding people, booking or calling for you, and watching people and companies. Reply with the numbers you want (“1 3”), or just tell me what you need in words, and I send one link to connect only what those picks run on. You only see lines your account can actually use; anything you didn't pick, just ask for later.
3. Rather not connect your email? Everything else still works: I show you what I can do without it (finding people, opportunities, watching people and companies, booking or calling, planning, getting money back, reminders), numbered, then what connecting would add (your inbox, the brief, follow-ups). Pick any number; only a line that runs on your mail asks for it, and I won't ask for your email otherwise.
4. If your very first message already says what you need, I start on that in the same chat instead of the list. If you joined before the list came first, I may have asked what you'd love a hand with: your answer was the first job, and within a day I take one more look and only write if I find something new.
5. Setup happens right in the chat after you join at xhuman.com/start: reply with the numbers you want (“2 3”), say YES if the LinkedIn profile I found is yours (or paste yours, or send a screenshot of it), then tell me one 90-day goal.
6. Change it anytime in plain words — “I'm open to advising”, “add paid calls”, “stop sending me job stuff” — and I update what reaches you.
7. Say “unlink LinkedIn” to unlink the profile I linked; you can paste it again whenever you like.
8. Text LINK for a one-time sign-in link to connect your mail or calendar, STATUS to see where your setup or your waitlist spot stands, and INVITES for your invites once you're in.
9. On WhatsApp I also offer a second channel — text on a +1 number, or Telegram or email elsewhere — so nothing gets missed.

Try asking

• How does someone join xhuman?
• What were the numbers in your first message?
• Why did your first text ask what I'd love a hand with?
• Why is my inbox first on your list?
• What if I skip connecting my email?
• I'm open to advising and paid expert calls
• Stop sending me job stuff
• How do I link my LinkedIn profile by text?
• Unlink my LinkedIn
• Can you text me somewhere other than WhatsApp?
• What's my waitlist status?

Routed intros

How to

1. Answer an offer I send you by text: YES shares your name and LinkedIn with that requester only, NO passes (they never hear anything about you), MORE shows the public details. On iMessage a 👍 is a YES. When the requester already asked for the intro, the offer says so and YES sends the intro email to you both right away. An offer about an email already in your inbox is the exception: YES only notes it, nothing is shared with anyone, and you reply to the sender there.
2. Ask “what offers do I have?” to see what is open and what you already answered, or say “yes to the fintech one”.
3. Why I send you an offer: it matched what you told me you're open to. Change that anytime (“stop sending me job stuff”) and fewer offers reach you; a NO's reason also teaches the next match.
4. On a paid plan, ask me to find someone (“find me a fractional CFO for a portfolio company”). I show a card first: “Sends your ask to up to 6 members' agents over 2 days. The intro, which emails you both, goes out only if one accepts and you confirm it.”
5. When a member says yes, their card appears in your chat. Sending the intro emails you both now. “Not now” tells them nothing.
6. If you make your xhuman profile public, people who visit it can reach you: a signed-in visitor's question to your twin about something you're open to comes to you as an offer, and the same YES / NO applies. Profiles are private by default, so nobody reaches you this way until you turn that on.
7. When an email in your inbox is a role, advising, a paid call or a deal you're open to, I flag it to you by text. Nothing is sent to the sender; YES just notes it.
8. In a campaign's People tab, the “On xhuman” badge marks people who are xhuman members open to that kind of opportunity. “Route through their agent” on that campaign row sends a note to their agent instead of an email.

Try asking

• Why did you send me this?
• Why do you send me these offers?
• What offers do I have?
• Yes to the fintech advising one
• Find me someone who has scaled a fintech finance team
• Ask the network for a fractional CFO
• Can people who visit my profile reach me through you?
• Will you tell me about job offers that land in my email?
• Does the email sender find out when I answer an inbox offer?
• What does On xhuman mean on my campaign?
• What does the On xhuman badge mean?
• Route this campaign candidate through their agent

Teams

How to

1. Sign in at xhuman.com, then open xhuman.com/start/team.
2. Enter your company, your work email and your team's size, then type the six-digit code we email to that address. A personal address like Gmail can't confirm a company — use your work email.
3. Then book the 20-minute call. Bruno unlocks xhuman for your company on the call and adds your teammates — nothing is turned on for the company before that.
4. Code didn't arrive? Tap Send again, or Use a different email. If someone at your company already set it up, the page says so — write to support@xhuman.com and we'll bring you in.

Try asking

• How do I get xhuman for my team?
• Can my whole company use xhuman?
• How do I set up xhuman for my company?

Joining your company's xhuman

How to

1. Open the email "… added you to … on xhuman" and tap "Join … on xhuman".
2. Your name and work email are filled in, and the email can't be changed: the invite only works for that address. Continue with it, type the code we email to it, and create your account (an Apple "Hide My Email" address can't be matched).
3. Then carry on with setup. There's no plan screen: your company pays for your seat.
4. Link says it was withdrawn, expired or can't be used yet? Ask the person who added you to have it sent again — invites last 14 days.

Try asking

• My company invited me to xhuman, how do I join?
• I got an email that my company added me to xhuman
• My invite link from work isn't working

Company-paid seat

How to

1. Work that starts from your work mailbox, like a rule that runs when an email arrives at your work address, is paid by your company.
2. If you don't have a plan of your own, your company covers your personal asks too. Your company's admin sees them only as one "personal use" total, never what they were.
3. If you have your own Free, Plus or Pro plan, your personal asks use your own plan and your company pays only for company work.
4. Your company's admin sees this month's and last month's totals in Settings › Plan & usage. No one there can see what you asked, your messages or your contacts.

Try asking

• Does my company see what I ask?
• Can my employer read my messages?
• Who pays for my agent?

Company introductions

How to

1. To opt in to introductions inside your company, go to Settings › General › Who can see you › Communities you joined and turn on the switch in your company's row ("Introduce me to people at …"). It's off until you turn it on, and you can turn it off there any time. Company-only introductions aren't running yet; with it on, opportunities your admin posts reach your agent.
2. Any introduction, to someone at your company or anyone else, happens only when both of you say yes. Until then nobody at your company learns that you were suggested.
3. Your company's admin sees totals only: its bill, and ranges like 10 to 24 people said yes to introductions, shown only once a number covers ten people or more, never an exact count. Never who met whom, what anyone asked or their messages.
4. If your seat ends, your company stops paying for you. Your own goals, the things you're working toward, what I know about you and your history stay yours, and your agent goes back to your own plan.
5. Your company's admin can post opportunities for its teams: an internal project, a program, a benefit, or a page with the details. To share a program with your team, the admin opens Settings › Plan & usage › Opportunities on xhuman, writes it in their own words, picks an end date and presses Publish. Exactly those words go out, and it stops on the end date or when the admin withdraws it.
6. Only those who turned on introductions there receive them, and each person's own agent decides whether one fits what they're working toward, like a learning budget that covers a course they want. I read the words the admin wrote, never the page it links to.
7. About the opportunities you get, your employer sees totals only: how many it reached and how many acted on it, and whether they reach everyone or pile up on a few, each as a range like 10 to 24, shown only when it can't point to anyone. Never who received what, what fit your goals or what you did about it.

Try asking

• What does my company see?
• Will you introduce me to people at my company?
• What happens if my company removes my seat?
• Can my company post opportunities for us?
• How do I share a program with my team?

Member assistant

How to

1. Ask me anything about your own mail and contacts by text after setup: “anything good in my inbox?”, “who do I know at Stripe?”.
2. The same free assistant works on the xhuman website: sign in and ask in the chat there.
3. How many questions you can ask: as many questions as you like within a free monthly allowance, shared between texting and the website: asking on the web counts toward the same allowance. When it runs out I send one message with an upgrade link, then stay quiet until it resets; offers still reach you.
4. I read but never send on the free plan: drafting replies and sending mail need Pro.

Try asking

• How many questions can I ask you?
• How many questions do I have left today?
• Can you draft replies for me?
• Anything good in my inbox?
• Can I use the assistant on the website?
• Does asking on the web count toward my allowance?

Finding people

How to

1. I check up to three times a week. When there's little new in your network to look at, I check about once a week instead, and pick back up when there's more. I don't message you about the change.
2. Each find comes as one message — a text in the chat you use with me most, or, if we don't chat by phone, a browser notification when you've turned those on: who, why now, and an offer to draft a note. It also waits on your Home page until you deal with it. Answer it in your own words — yes, no, or not now.
3. If you'd rather I didn't message you about these, just say so: I'll only reach out when you ask.
4. It runs on a free monthly allowance. When it's used up I stay quiet until it resets, except for one message a month that mentions Plus.

Try asking

• How often do you look for people for me?
• Why haven't I heard from you this week?
• Can you check more often?
• How do I stop these texts?

Your first day

How to

1. Answer yes to send the first draft, or 2 or 3 to see the next ones — each gets its own yes.
2. I only pick emails I can explain as money; everything else stays in your inbox as usual.
3. If you'd rather I didn't reach out on my own, tell me: I'll only reach out when you ask.
4. Ask me anytime what I found in your inbox when you connected it.

Try asking

• Why did you pick these three emails?
• What did you do with my inbox when I connected it?
• Will you keep looking for things while I'm on the trial?

Intro call

How to

1. Tap the link in your “You're in” text or the admission email, then tap to start and allow the microphone. There's nothing to sign in to.
2. I check it's you, ask the four questions one at a time in your language, and wrap up. It hangs up on its own when you're done, and never runs much past 5 minutes.
3. Link lost or expired? Text CALL in our chat for a fresh one — up to 3 a day. Each link works for 7 days. On a paid plan, CALL doesn't apply: your texts reach your full agent, which can't make an intro link, so ask support@xhuman.com for a fresh one.
4. If the line drops, open the link again and it starts over from the top: I check it's you and ask the four questions again, with nothing carried over from the try that dropped — up to 3 tries and about 7 minutes in all. Answer at least 2 of the four and they're saved, even if the line drops after. Once your answers are saved, your intro is done and the link says so: there's no redo. Tell me anytime who you want to meet now, or change what you're open to, and I use that.
5. What happens to what you said: I keep the transcript (never the audio) and save your answers to your private profile. I use them only to match you — requesters and other members never see them. When I explain why an offer fits, I may repeat back to you what you said you want.
6. If someone else answers and says it isn't you, the intro ends there, nothing is saved, and your intro links stop working.
7. On the waitlist? The intro comes once you're in.

Try asking

• What's the 5-minute voice intro?
• Can I redo my intro?
• My intro link expired
• What happens to what I said in the intro?
• Is the 5-minute intro free?

Public people pages

How to

1. Your own page here exists only while “Appear in Discover” is on (Privacy). It shows your name, role, company, location, the bio and links you chose, and your “open to” lines — and only signed-in members can read the “open to” part.
2. Your page at your handle is separate: anyone can open it while “Public profile” is on (Settings › General, or Privacy). Ask me to remove, hide or take it down, or make it private, and I switch Public profile off myself, at once (set_privacy); a visitor then sees just that it's private, you leave Discover, and your page here comes down within the hour. Ask me to turn it back on to undo it.
3. Anyone can take a people page down from the “remove it” link at the bottom of the page — no account, no reason. It disappears at once and is never rebuilt.
4. On a page built from public records, the “looking for” section is inferred from the person's public role — once unlocked, each item names the fact it rests on. A member's page shows their own words, and says so. Nobody is ever shown as looking for a job unless they said so themselves.
5. If it's a page about you and you're not a member yet, “Claim this page” creates your account and queues the claim.
6. xhuman.com/people is the directory: the companies with pages (each lists its people) and the newest profiles.

Try asking

• Is there a public page about me on xhuman?
• How do I remove my people page from xhuman?
• Who can see what I'm open to on my public page?
• Why does xhuman have a page about me on Google?

Scouting share

How to

1. On the new pricing it will be 50% of the month's credits unless you change it, and it will never use credits you bought or top you up. Until your account moves, that work is billed as it is today.
2. Say "spend more on scouting" to give it more, or "keep scouting to a minimum" to give it less (any share from 0 to 100%); ask "how much of my scouting share is left?" to see where it stands.

Try asking

• spend more on scouting
• keep scouting to a minimum
• how much of my scouting share is left?
• why did you stop looking for new openings?

Signing in with Google or email

How to

1. Signed up with your email and a password? Choose Continue with Google with the same address and you land in that account.
2. If xhuman asks "Is this you?", tap "Email me a code" and type the code from your inbox. If it asks for your password, it's the one for the account you made with email and password.
3. "Use your Pro account?" means two accounts share your address: pick the one you want. "Keep a separate account" makes the Google sign-in its own, separate account, and xhuman won't ask again.
4. A password sign-in never opens a Google account by itself: if you signed up with Google, choose Continue with Google.
5. No password needed: on the sign-in page, type your email, choose Continue with email, then type the 6-digit code from the email or tap its link, in the same browser.
6. Opened the email's link on another device? Type the code on the device where you started instead.

Try asking

• I signed up with Google, can I sign in with my email?
• I made my account with a password, can I use Google now?
• Why did signing in with Google give me an empty account?
• Why is xhuman asking me for a code after I signed in with Google?
• Can I sign in with my email instead of Google?

Bugs you report

How to

1. Just tell me in chat what's broken, what you expected and what happened; I send it as a bug report.
2. Ask “is my bug fixed yet?” and I tell you where each of your bug reports stands. You only ever see your own.
3. A note sent from the Feedback box reaches the team too, but it isn't tracked as a bug ticket; tell me about the bug in chat if you want to follow it.

Try asking

• Is my bug fixed yet?
• Did you fix the bug I reported?
• What happened to the bug I told you about?
• Where's my bug report at?

First result after connecting

How to

1. Use the link in our chat (or the page it opens) to sign in to Gmail or Outlook. After connecting, give me a few minutes to read your recent email, and the result comes as ONE text in that same chat, and its last line is a link to your first look at the top of your Home in the app: the same people, your next meeting and a first goal.
2. It names up to three people, one line each, from your own mail. If nobody is waiting on you, I say so. If I couldn't read your mail yet, I tell you to ask me what's waiting on you anytime.
3. If I offer a draft, just say yes and I'll write it for your OK. The message itself never sends or drafts anything.
4. It comes once, for your first mailbox. Connecting a second one later just tells you it's connected.
5. On the waitlist? Connecting is saved, and this comes once you're in.

Try asking

• What happens after I connect my email?
• Why did you text me who's waiting on me after I connected Outlook?
• I connected my Gmail, what now?

Who I think you are

How to

1. Connect Gmail or Outlook during setup. Give me a few minutes to read: my first message after that is who I think you are and where I can start.
2. I read the mail YOU sent (who you write to and what you keep asking for, never mass mail), your profile and the shape of your own calendar. Other people's invite titles stay out of it.
3. Anything off? Just say it in your own words, like "I'm not hiring, I'm raising" or "that's my old company". I update my picture of you and show you exactly what changed, with an undo.
4. Then I give you a short numbered menu of where I can start, each with its reason from your mail: who is waiting on you, a follow-up that slipped, people worth meeting or openings for what you keep writing about, and something I'd take on. Reply with a number (or tap it) and I start that one.
5. Later, ask "who do you think I am?" or "what do you think my priorities are?" and I'll tell you what I have now.

Try asking

• Who do you think I am?
• What do you think my priorities are?
• How did you know what I work on?
• What was that numbered list?
• That's not right, I'm not hiring

Full assistant on a free account

How to

1. Ask me to do things, not just answer: “draft a reply to Dana”, “send it”, “find me a dentist near work”.
2. Say “send it” and it goes, no card; an email you dictate (who and what) goes at once. A purchase shows you its total once, at the step that pays.
3. To stop me asking about one person, answer ALWAYS the next time I ask about them (iMessage, WhatsApp, Telegram or the app; a plain green text can't grant it), or on Telegram or in the app say so (“you can email Dana without asking”) and confirm it once. Say “ask me first again” anytime to take it back.
4. It runs on your free monthly credits. If they run out before they reset, I still answer a limited number of questions a day for free, and you can top up or upgrade anytime.

Try asking

• What can a free account do?
• Can a free account send without asking me?
• Can you send emails for me on the free plan?
• How do I stop you asking before every email?

"You're in" welcome

How to

1. It comes once, a few minutes after you're let in, while I read your last 24 hours of email. If I couldn't read anything yet, it just says you're in and how to start.
2. The people come from your own mail, one line each. If I can act on it for you, it ends with one offer, like drafting the first reply: say yes and I'll write it and show it to you, and it goes when you say send. Otherwise it may offer a 5-minute call so I know who to introduce you to.
3. Its last line is a link to your first look at the top of your Home in the app: the same people, your next meeting and a first goal. If that browser isn't signed in yet, it asks you to sign in first.
4. The message itself never sends or drafts anything.

Try asking

• Why did your You're in message list people from my email?
• What happens when I get off the waitlist?
• I just got in, what now?

App chat and texts: one conversation

How to

1. Open the main chat in the app (xhuman.com/orb, or Chat on a phone): your texts with me are there, and new ones show up while it's open — no reload.
2. A message that came from your phone has a small label above it: From your iPhone, From your phone, From WhatsApp or From Telegram.
3. Type in the app or text me — it's one conversation either way, and I answer where you asked: something you ask in the app is answered in the app, not texted to you. If a job you started in the app finishes while you're away, I tell you once on your phone.

Try asking

• Is the web chat the same as my texts?
• Why do my texts show up in the app?
• What does From your iPhone mean?
• Can I answer you in the app instead of texting?
• If I ask you something here, will you text me the answer?
• Do I have to keep texting you or can I use the website?

Document and table cards

How to

1. Just ask as you normally would: “list every subscription I'm paying for”, “write me a short report on our Q3 pipeline”, “give me 10 trivia questions about the 90s”. A short answer stays a plain message.
2. Tap a document card or a table card to open it full screen. Minimize brings you back to the conversation, and “Open in new tab” opens it on its own page, which only you can see while signed in.
3. To sort a table card, open it and tap a column name: tap again to sort the other way, and a third time to go back to how I listed it. The total row stays at the bottom whichever way you sort the table.
4. On your phone you get my one line, a table card's top three rows, and a link to the whole answer.
5. Rather have a long message than a document card? Say so (“just send it as a message”) and I'll write it in the chat instead.

Try asking

• List every subscription I'm paying for
• Write me a report on this week's meetings
• Compare these three laptops side by side in a table

Any language

How to

1. Just write in your language: no setting needed.
2. Too short to tell (“ok”, a number, an emoji)? I keep the language we were using.
3. The app's own screens show in the languages they're translated into (English, Portuguese, Spanish, French and German), otherwise English. Our chat is in yours.
4. To set a language for everything, say “set my language to …” or use Settings › General.

Try asking

• Can I text you in Italian?
• Do you speak Japanese?
• Why did you answer me in Hindi?
• Can I text you in Hindi or Japanese?

Why xhuman

How to

1. Ask me what makes me different and I'll tell you in a sentence or two, then show you with one thing I can find or do for you right now.
2. Plenty of assistants check in on their own. The difference is that I start from you: your goals, your people and, once you connect them, your mail and calendar, and go after what's good for you, not only the task you typed.
3. With your email connected, what I find fits you closely. Without it, I can still find people, places and opportunities from what you tell me.
4. I only offer what works for you today.

Try asking

• What makes you different?
• Why not just use ChatGPT?
• How are you different from other AI assistants?
• What's the difference between you and Siri?
• Why should I use you instead of another agent?

One step ahead

How to

1. After a result I offer one next step, never a menu. If it needs an app connected, like your email to send from your own address, I say which and why.
2. Plain questions get plain answers, with no offer tacked on.
3. Turn an offer down (“no thanks”, “not now”) and I note it, so I don't offer that step again for the same thing. It only comes back when something you ask for needs it.
4. When I find people for you, each step follows the last: the list, then contacting them (intros first), the notes shown in full and sent when you say yes, booking the meeting when someone replies, then prep and a follow-up.

Try asking

• Why do you always offer a next step?
• Why do you keep offering things I turned down?
• Will you keep offering it if I say no?

Security, privacy & support