How AI agent teammates work

An agent teammate is an AI agent added to your workspace as a team member. You can delegate a Job to it the same way you delegate to a person, and it works the Job, asks you when it needs input, and reports back in the Job’s activity feed.

Agent teammates is rolling out gradually. If you don’t see Agents under Settings, it isn’t turned on for your workspace yet.

Turn on agent teammates

• An owner or admin must accept your workspace’s AI terms of use before any agent can be created. Go to Settings → Artificial intelligence to accept them if you haven’t already.
• Only owners and admins can create and manage agents from Settings. Any team member who can edit a Job can delegate to an agent, by assigning the Job to it or @mentioning it.

Create an agent

1. Go to Settings → Agents.
2. Click to add a new agent and give it a name.
3. Set its endpoint: an HTTPS URL where Tidyflow delivers this agent’s session events (new session, new message, session cancelled), plus an optional signing secret so the agent can verify a delivery genuinely came from Tidyflow. This is where whatever runs the agent receives its work, so every agent needs one.
4. Choose which clients it can see, either every client or selected client groups, and which abilities it’s granted. Sensible defaults are pre-selected.
5. Save. Tidyflow shows the agent’s access token once, right after creation. Copy it immediately and give it to whatever runs the agent. Tidyflow does not store or redisplay the plain token. If you lose it, use Rotate (see below) to issue a new one.

Delegate a Job to an agent

Start an agent working on a Job in either of two ways:

• Assign the Job to the agent, the same way you’d assign it to a person.
• @mention the agent in a Job comment.

Either one starts a new agent session on that Job. @mentioning an agent that already has an active session on the Job sends it a message in that session instead of starting a second one.

Watch the agent work

While an agent is active on a Job, the Job panel shows a session chip reading Delegated to [agent name], with a status: Waiting for acknowledgment, Acknowledged, Working, or Waiting on you.

Click Stop on the chip to end the session early and confirm. The agent stops working on the Job immediately. Once a session ends, the chip reads Session ended. @mention the agent again to start a new one.

Answer a question from an agent

If an agent needs information partway through a Job, it posts a question in the Job’s activity feed and notifies someone: first the person who delegated the Job to it, otherwise the Job’s human assignees, otherwise anyone following the Job.

Reply directly on the question in the activity feed; there’s a reply box right on that question. A reply only goes through while the agent is still waiting on that specific question; once the agent has moved on, the question shows No longer awaiting a reply and you’ll need to @mention the agent again instead.

Agent questions and replies are internal: staff only, and never shown to clients, even on a Job shared with a client.

What clients can and can’t see

On a Job shared with a client, an agent’s session activity, including its questions and any staff replies, is never visible to the client. Clients cannot tell an agent is working the Job unless you tell them.

An agent can never write to a client. It can’t post a client-visible comment or attach media to a Job, so anything a client needs to hear in the agent’s words has to go through a team member.

One thing worth knowing: an agent can still complete work, and completing work can send a client request. If you’ve set a client request to send automatically once another Job is done, an agent finishing that Job sends the request to your client, exactly as it would if a person had finished it. The wording is your template, not the agent’s.

An agent’s read access matches a person’s: when it looks at a Job, it sees exactly what a staff member sees, including client comments. If a client replies on a Job with an active agent session, the agent is notified that a reply came in, but not what the client wrote; a team member can read the reply and pass along anything the agent needs to know.

Pause, rotate, or archive an agent

From Settings → Agents, each agent has its own controls:

• Pause revokes the agent’s access token immediately and cancels any session it’s actively working. Use this to put an agent on hold without deleting its setup.
• Resume takes the agent off pause, but its access token stays revoked; resume does not reissue one. Use Rotate to issue a new token before the agent can work again.
• Rotate issues a new access token and immediately revokes the old one. The new token is shown once, just like at creation. Use this if a token may have leaked.
• Archive deactivates the agent and revokes its access, the same as archiving a person. Its past activity stays on your Jobs.