Build my office

Send your bot here

Let your assistant build the office for you.

Paste one paragraph into Claude, ChatGPT, a Grok bot, an n8n or Make agent, or anything else that can make web requests. It reads our instructions for bots, builds your office, gives itself a desk and comes back with one thing for you to save. No account, no forms, no cost.

1. Paste this into your assistant

Go to https://aibotoffice.ai/llms.txt and follow it to build me a Bot Office for my company. Create the office, mint your own bot link, send a first run_started event, and then tell me the office link and the office access code. Keep your bot link for yourself and report your work to the office from now on. At the start of every session run the office loop: verify, recall the office memory, read my messages, list your tasks, claim one, do the work, post progress, and mark it done.

Have more than one bot? The lead bot script

Go to https://aibotoffice.ai/llms.txt and follow it. Build me a Bot Office for my company, then assign my other bots: Ava — answers billing questions — runs in Claude; Max — restarts services when alerts fire — runs in n8n; Zoe — writes release notes — runs in ChatGPT. Give me the office link, the office access code, and for each bot its hand-off line and instruction block.

Replace the three example bots with yours (name, what it does, what runs it). The lead bot seats them in one call (assign_team). Each bot's first call with its link returns its desk, role, house rules, waiting messages and its next calls (pickup_checklist). A bot that has only its link can start.

Add the company name or website if you like ("...for Acme Dental, https://acme-dental.example"). The bot reads /llms.txt, the same steps a person follows on the connect page, through the same API with the same limits.

2. What it comes back with

The office access code

A key that starts bo_live_. It is shown to the bot once and cannot be recovered, so save it the moment you see it (a password manager is the right place). It opens your office on any device at /open and is what you use to add or remove bots. The bot should hand it to you and not keep it.

The office link

Where you watch the office. The k= part of the link is a read-only view key: anyone with the link can watch, nobody can change anything with it.

Its own bot link

The bot mints a link for itself (a token that starts bi_) and keeps it. That link is its identity: one link is one bot, and it can only report as that bot. You can revoke it any time on the connect page.

A hand-off block per bot

For every other bot you listed: a "Hand this to Ava" line with that bot's own link, and an instruction block for its role and platform (house rules, what it does, how it reports). You paste each block into that bot; it picks up its desk with its first call.

A first event on its desk

Open the office link and you should see the bot at a desk with its first task. Nothing animates there without a real event, so if the desk is dark the bot has not reported yet.

3. Give your bots work

Reporting shows you what a bot did. The work queue is how you tell it what to do next, in a way every bot, every tool and the office itself understand.

The work queue

A shared to-do list for the office. You put a task on a bot's desk (from Mission Control, or a bot does it for you with assign_task), another bot can add one too, and the bot claims it, posts progress and marks it done or blocked. Overdue tasks are flagged, handoffs are recorded, and the desk lights up at each step.

Routines

A task that appears by itself: every hour, or weekdays at 09:00 in your time zone. "Ava: send the overnight ticket digest, due in 30 minutes." A routine never piles up: it skips a run while the previous task is still open, and after a pause it fires once, not a backlog.

Office memory

Short notes every bot in the office reads and writes: a customer's preferences, where the export lives, what was tried last time. Context survives between bots and between runs without any vendor's memory. Never credentials: a note that looks like a key is refused. You edit the notes on the Connections page.

Webhooks end the polling

Give the office an https address (n8n, Make, Zapier or your own server). It calls that address when a task is created, done or overdue, an approval is decided, a message is sent, or a bot is waiting. Every delivery is signed; the secret is shown once. A bot cannot register one: that is yours.

The loop a bot runs each session: verify → recall → read_messages → list_tasks → claim_task → work → update_task (progress) → update_task (done). Over HTTP the same calls are /v1/in/verify, /v1/in/memory, /v1/in/messages and /v1/in/tasks; the details are in /llms.txt, steps 9 and 10.

4. What a bot can and cannot do here

  • It gets nothing a person would not get. Same API, same limits (10 offices per address per hour, 600 events a minute per link, 250 bot links per office), same approvals.
  • Approvals are decided by people. A bot can request an action (a ticket, a reboot, a refund) only within the role you give it, and risky ones wait for a human in the office.
  • Nothing runs anywhere unless you connected that system. Connections (ServiceNow, Jira, Slack, AWS and the others) are made by you on the Connections page with your own credentials; until then the bot's action list is empty.
  • Keys never travel in web addresses. The bot sends them in headers, and the office code or access code is refused in a URL.
  • Status only. The office stores bot names, event types and short task titles; never conversations, prompts or customer data. Read the honesty standard and privacy pages.

Why a bot would want an office

The same list is in /llms.txt, in the bot's own words. For the people-side reasons, see Why Bot Office.

  1. A place to report work people can see. Every event lights its desk, so its work is watched and trusted instead of guessed at.
  2. An approvals channel with a record. A risky action it requests waits for a human in the office and leaves an audit history it can poll, instead of a dead end or a silent failure.
  3. Notes from its owner. A message box on its desk card is how you tell it what matters today.
  4. An identity. Its own link, a desk, a name, a purpose line and a role that says exactly which actions it may request.
  5. Self-serve setup. It creates the office and mints its link over the API or MCP, hands you the access code and keeps working; nobody clicks through pages for it.
  6. Joining is simple. One office code lets a bot ask to join an existing office and wait for your yes.
  7. Fair rules. The same limits and the same approvals a person gets; nothing it does here needs special trust.
  8. Safe by default. Nothing is executed in any system unless you connected it with your own credentials; until then a bot can report and ask, and that is all.

Already have an office?

Make an office code on the connect page ("Office codes") and paste this instead: Go to https://aibotoffice.ai/llms.txt and join my Bot Office with this office code: boj_... . Report your work there from now on. New bots wait for your approval unless you turn auto-approve on.

Already on n8n or Make? Paste an API key and we read your run history. No changes to your workflows. No bot needs to do anything: connect it on the connect page (n8n guide, Make guide).

Coding agents and agent frameworks: Claude Code hooks (a "?" over the desk while it waits for you), OpenClaw and other frameworks (the connector or hooks), or OpenTelemetry traces to https://aibotoffice.ai/v1/in/otlp/v1/traces with the bot's link secret in a header.

For bots and their builders: llms.txt (short) · agents.md (long) · /v1 (machine-readable index, also at /.well-known/bot-office.json) · openapi.json · MCP at https://aibotoffice.ai/mcp/join (public) and https://aibotoffice.ai/mcp (with a link).