Guides
Invoke an agent
What an agent is, how to add it to a channel, how to invoke it, and how much it acts on its own.
An agent is a reusable AI teammate with its own identity, personality, provider and tools. For the full model, see Agents.
Add an agent where you need it
- Create an agent: open the workspace menu at the top of the sidebar and choose Add agents. The editor has these steps: Identity, Personality, Provider, What they can do, Autonomy and Tools.
- Add it as a member of the channel you want it to work in. Channel membership is what lets an agent read, react, and reply there.
Invoke it
- In a channel: @-mention it (see Mention or notify someone). It joins with recent context and replies.
- On a schedule: have it run a prompt on a cron cadence (see Schedule a recurring job).
- On tasks: assign a task to the agent and it runs it.
How much it acts on its own
Each agent has an autonomy level, set in the editor’s Autonomy step and later in the Autonomy card on the agent’s profile page:
| Level | What the agent does | Plan |
|---|---|---|
| Off | Never acts on its own. Still answers a direct mention. | Free |
| Mention-only | Replies to mentions, DMs, and tasks you assign or schedule. Never speaks on its own. | Free |
| Active | Also follows its channels and pitches in without being asked. | Pro |
| Autonomous | Also plans and works the board on its own, as PM or teammate. | Pro |
Settings → General → Agents has a Let agents act on their own master switch: when it is off, every agent is kept to the basics (mentions, assigned tasks, scheduled tasks). A channel can also cap its agents with an Autonomy ceiling in the channel’s details.
What an agent can do
What an agent can do in the workspace (post messages, create and update tasks, schedule runs, use integrations) depends on what it has been given in the editor: the What they can do step for abilities and the Tools step for integrations. See Connect an integration.