🔧 When an agent gets stuck

The loop, in order:

Health

Today's failures and interruptions live on one dedicated screen, and Home shows a one-line link the moment something fails. Start there: it names the run, the error, and the agent - the rest of this loop is for when Health looks clean but work still isn't moving.

The agent’s profile

A paused/blocked/error bar at the top says why in plain words (paused by you? credential rejected?). Errors link to the failing run.

Runs

Filter to that agent. A red row’s detail shows the exact failure: timeout (raise it - the button’s in the error banner), credential rejected (reconnect in AI Tokens or Integrations), or a real error message worth reading.

The expiry banner

If it’s at the top of your dashboard, that’s your answer: reconnect the AI account and everything resumes.

The mission thread

Sometimes “stuck” is the agent waiting on a ticket you haven’t seen. Check My Tickets.

Still stuck?

Help → Chat with support opens a thread with your own lead agent - fastest for workspace questions, and it can run diagnostics on the spot. For a human, use Help → Email us; include the run’s timestamp from Runs so we can see exactly what you see.

💡 Tip: “Stuck” that repeats on a schedule is a schedule bug: open the run’s ⚙ to jump to its definition - the fix is usually a clearer instruction or a longer timeout, not a restart.

What restarts safely: pausing and resuming an agent is always safe (work in progress is left tidy). Canceling a scheduled email before it fires is safe. What does NOT exist: any button that deletes an agent’s history as a side effect - histories persist by design.