Skip to main content

​
What this does

These commands show fleet activity and change lifecycle state. Every command uses the active workspace. The dashboard uses your chosen fleet name in Your fleets and the conversation header. Open a card to return to that fleet.

​
Before you begin

Install a fleet. Copy its identifier from agentsfleet list.

​
Steps

  1. List fleet identifiers.
  2. View recent activity. <FLEET_ID> comes from the list output.
  3. Send a manual message.
    Approve the fleet’s integration grant during setup to allow repository access. That approval lasts until you revoke it. Custom approval rules can pause individual runs; review those requests with agentsfleet approvals. In the dashboard chat, a reply appears as it arrives. You can send another message while a run is active; the new event waits its turn and keeps its own place in the conversation. Reasoning and tool activity are shown separately from the final reply. While a reply runs, the chat keeps your place. Your new message moves to the top of the panel, and the reply grows below it. If you scroll back, a turn someone else sends does not move you. Click Jump to latest to follow it. To change the fleet’s model, choose a model in the chat composer. The saved choice applies to future chat, webhook and scheduled runs. Saving keeps your unsent message and private setup edits. Model changed appears in Chat and Activity after the save. Choose Run details to open usage and duration details. Press Escape or click outside the details to close them.
  4. Stop new work while keeping the fleet resumable. In the dashboard, open the fleet header’s … → Stop and confirm.
  5. Resume the fleet.

​
Verify it works

Waiting counts approvals that need your decision. When any fleet is waiting, the client adds this line:

​
Common problems

UZ-AGT-009 means the fleet does not exist in the active workspace. Check the workspace and fleet identifier. A killed fleet also returns UZ-AGT-009 when you try to rename it or change its status with a current version. Create another fleet to resume work. For API lifecycle updates, send the strong entity tag (ETag) from the fleet body’s etag field in If-Match. Compression may weaken the response header. A missing If-Match returns 428; an outdated value returns 412, including for a killed fleet. UZ-AGT-012 means the fleet cannot accept a message in its current state. Resolve the cause or resume a stopped fleet. If a model save fails, the composer keeps the last saved choice. Choose Retry for an unconfirmed save, Reload choices to refresh available models, or Reload fleet when another save changed the setup. A run with no reply may be waiting for approval. Run agentsfleet approvals list to review the request. A message with no answer after 30 seconds reads Couldn’t confirm this message was sent. Click Resend. It carries the same operation_id, so the fleet runs the message once. A failed reply keeps any partial text in the conversation. Choose Try again when available to send the original request again. A retry creates another run and can incur another charge. Before retrying a request that changes an outside service, check whether the action already completed. A new run can repeat completed actions. Open View details for diagnostics. Event details also offer Copy diagnostic, which includes the event identifier and recorded failure cause. A lifecycle command sends a write again only when the connection was refused, the name did not resolve, or the server declined it. A reset after sending or a timeout surfaces the error instead, because the server may have processed the write. Check agentsfleet doctor, then run the command again.

​
Live chat connection

When live updates stop, Chat reconnects. A stable connection has 25 seconds to recover before Chat shows a warning. A stalled connection attempt retries after 30 seconds. An open connection retries after 45 seconds without activity or a server heartbeat. Chat makes five fast retries, with increasing delays capped at 15 seconds. Then it shows Live updates are temporarily unavailable. We’re reconnecting automatically. It keeps retrying every 30 seconds. Click Retry now to reconnect immediately. Returning to the tab or restoring the network also starts recovery when the connection is missing or stale. After reconnecting, Chat fetches recent saved activity. Missed reply fragments and tool updates may remain missing.

​
Remove or undo

Stopping is reversible with resume. Killing is permanent but keeps the fleet records until deletion. In the dashboard, choose … → Kill from the fleet header. Confirm the permanent action in the dialog. Deleting removes all fleet records and cannot be undone. The client has no dry-run option for either command.