Know what the queue will do.

How do priorities and dependencies work?

The current worker uses the order of the task manifest. It does not schedule a dependency graph or learn priorities from outcomes. Put only independently ready tasks in that list. Check a prerequisite result before creating or authorizing dependent work.

Customized task graphs and outcome-based ranking are product work, not active features of this preview. A task’s priority never grants permission to run it.

Does AgentAFK run a cloud worker for me?

No. You run the queue on your own computer or cloud machine. Hosted execution and cloud supervision are not available from this site. Downloading a manifest does not start a worker or a schedule.

Can I use an inexpensive model?

Use a supported configured provider that fits the task. Start with a short assignment and check its output. The queue does not guarantee task completion, lower spending or measured token savings.

What happens after an interruption?

The worker retains its call reservation and original deadline. Failed or uncertain outcomes hold further dispatch. Inspect artifacts, provider records and external effects before following the explicit reconciliation steps in the worker guide.

Why is the report file refused?

Use the JSON output of the queue’s --status command, up to 4 MiB. The viewer accepts overnight_queue_status/v2. It refuses malformed records and reservations that exceed the recorded budget.

Where is my data saved?

The site does not upload task paths or imported status files. Save your downloaded manifest yourself. Reloading the page clears the form. The worker’s output and state remain wherever you configured them on your machine.

Where can I get help?

Check the setup steps and service connection. Report a reproducible software problem in the public repository. Do not post credentials, private task files or customer data.

Operated by Baltor.AI. Postal contact: 1428 Bryn Mawr St, Saxton, PA 16678, United States.