Use a built-in agent

A built-in agent works from a thread. You tag it like a person. It gets the conversation and the page, then answers in the thread or works in the project’s repository.

Tag the agent

In a new note or a reply, type @ and pick an agent from the list. Write what you want done around the mention, then send.

The agent receives the thread, the page, and the element you pinned. Claude and Grok also get screenshots. The other agents get the same details as text.

If an agent is not in the list, the project has not set it up, or you are not allowed to tag agents. Ask a project owner. See Agents.

Reviewers who only comment cannot tag agents.

Follow the task

The agent’s reply appears as a task card. While it runs, the card shows its status, progress, and a stop control. When it finishes, the card shows the answer, and for repository work, the pull request or commit.

Add context while it works

Reply in the same thread and tag the same agent again.

• Claude, Cursor, Devin and Jules take your reply into the running task.
• Codex, Copilot and Grok do not. A new mention starts a new task, so check the card first.

Stop a task

Select stop on the card. Any workspace member can stop a task, not only the person who started it. Reviewers who only comment cannot.

If a task fails or stops

A stopped task was canceled on purpose. A failed task hit an error, and the thread says it could not finish.

A project owner can read the error on the agent’s card under Projects → your project → Agents. Common causes are an expired or used-up provider credential, a provider limit, or missing repository access. Fix the setup before starting another task. An agent with a recorded failure may drop out of the list until it runs successfully again.