AI agents: Billy and your own agents

Billy is your office’s AI colleague. You can add more agents next to it, each with its own name and job, and mention them like employees.

An agent never acts with permissions of its own. It sees only what the person asking may see, and changes nothing unless that person approves. An automation it drafts stays a draft until they turn it on.

Ask Billy

• In a task: write a comment, type @ and choose Billy (or another agent, marked AI). Ask what you need, for example “What is still missing from this client for the annual report?”.
• In team chat: mention Billy in any conversation, even a group it is not part of.
• In a client conversation: mention Billy in an internal note.
• In a private chat: in Messages → Team, start a new conversation and choose Billy.

While Billy works you see Billy is working on it…, and you can Stop. Billy looks things up in the office’s data itself, with your access only.

The answer is yours first

In a task, a group chat or a client conversation, the answer appears in a card marked Only you can see this, because it may include something others there may not see. You decide:

• Post here adds the answer to the thread as Billy’s, with posted by and your name.
• Dismiss removes the card.
• How it got there opens a side panel with what Billy looked up on the way, and Token use: the total, input (new, read from cache and written to cache), output (with how much was reasoning), how many times Billy called the model, and which model answered.

In a private chat with Billy the answer goes straight into the chat.

Billy suggests, you approve

When you ask Billy to do something (open a follow-up task, mark a task done, send a client a message), it does not do it itself. The card shows Billy suggests with exactly what will happen: the task’s title, client and date, or the message, the address it goes to and the office channel that sends it.

• Approve makes the change once, as you and with your access.
• A message to a client is sent only after Approve and send and a second confirmation.
• Reject drops the suggestion.

Only the person who asked Billy can approve. A suggestion nobody approves expires after a day.

Billy drafts automations

Ask Billy for an automation, or a change to one, for example “@Billy every time a client becomes active, open a welcome call task for the client owner”. Billy hands the request to the automation drafting and answers with a link to the draft. The draft is yours: nothing runs until you open it, check it, try it and turn it on. Only owners and admins can have automations drafted. See Build an automation.

Turn on Billy

Owners and admins only.

1. Open Settings → AI agents.
2. Click Turn on Billy.

Billy appears in the list with the Main tag. It cannot be removed, but you can pause it.

Add an agent

1. In Settings → AI agents, click New agent.
2. Give it a Name. This is the name people mention.
3. Write a Job description: what it is for, what to pay attention to, and how the office likes things done.
4. Decide whether it May read sensitive data (see below). Off by default.
5. Click Create.

An office can have up to 5 agents, Billy included. To add another, remove one first.

What agents can read

The What agents can read section on the page shows where the line is. In every case an agent reads only what the person asking may see.

• Every agent: clients and contacts (names, tax ID, which for a private client is their ID number, email and phone), tasks, projects, team chats, client conversations, bookkeeping, reports and annual report worksheets.
• Only agents allowed sensitive data: ID numbers and dates of birth anywhere else (spouses, children), amounts and details read from documents and their change history, signatures, identity documents, clients’ answers to information requests, declared assets and debts, staff hours, attendance and payroll, and exports of every client or contact.
• No agent: passwords, connection keys, client links, other offices, and each person’s own settings.

To answer, what an agent reads is sent to the AI model. It is not used to train models.

Edit, pause and remove

• Edit changes the name, job description and access to sensitive data.
• Pause stops the agent: it does not answer until you Resume it. Earlier replies stay.
• Remove (not for Billy) takes the agent out of mentions. Replies it already wrote stay where they are.

When it does not work

• If the page says AI agents are not available yet, they have not been opened to offices.
• If your office’s plan does not include AI agents, the page says so. See Plan and usage.
• Members who are not owners or admins cannot manage agents.
• If the office used this month’s AI allowance, agents say so and stop answering until next month. Plan and usage shows how much is left.