AgentAFKPrepare a run ↗

Worker prerequisites

Prepare here. Execute on your worker.

The planner, manifest download and report viewer work in your browser. They do not install or start a worker.

Worker installation is unavailable in this preview. Only run a generated command if you already have a compatible, reviewed worker installed. You can prepare task lists and inspect existing reports without it.

  1. Check an existing worker

    Use a Linux, macOS or WSL worker you control. The command template expects its Python environment at .venv/bin/python and queue runner at tools/overnight_queue.py. Verify those files and their installation instructions first.

    .venv/bin/python tools/overnight_queue.py --help

    If that command is unavailable, stop at manifest preparation. Do not download or run a substitute without reviewing its source and permissions.

  2. Check your model access

    Use the provider configuration supported by your installed worker. Choose an endpoint you can reach, such as your own network service or Ollama Cloud. A local model download is optional.

    Keep keys on the worker, outside task files and this website. Check the provider’s spending controls before a model call.

  3. Write one short task

    State the objective, permitted inputs and changes, destination files, and checks for success. Include only work whose prerequisite results are already approved.

    A useful task file includes
    • Objective and expected output
    • Allowed files and effects
    • Time and provider limits
    • Checks that determine acceptance

  4. Prepare and inspect the queue

    Enter the absolute task-file path in the run desk. Download the manifest, save it in your checkout, and inspect the generated command before running it.

    Open the run desk →

  5. Review the result before leaving more work

    Check the worker’s status and output files: the output files confirm whether the task succeeded. Preserve the original run folder for interrupted or uncertain work.