Get started

The shortest path to a verdict is a log file and one command — no CI, no token, no model. Start there, then wire it into a pipeline.

Install uv, pip, the Docker image, or no install at all on GitHub Actions. First run Replay a saved log offline, then point it at a live pipeline. What the report looks like. Requirements Python or the image, a read-scoped token, and what it does on an air-gapped network.
On GitHub Actions there is nothing to install. One step, and it defaults to the run that triggered it — skip straight to GitHub Actions.

What "getting started" actually involves

  1. Install it — or don't, on GitHub Actions.
  2. Replay a saved log offline to see the report shape.
  3. Get a read-scoped token when you want it to fetch a live run itself.
  4. Add the job to your pipeline, where it runs only on failure.

Steps 1–2 need nothing but Python. Everything after that is about letting it reach your CI provider. An LLM is optional at every step — without one you get the full deterministic report, not a stub.