> For the complete documentation index, see [llms.txt](https://docs.perception.cx/perception/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.perception.cx/perception/enma-ai/autopilot.md).

# Autopilot

Give Autopilot a goal and it works turn after turn until the goal is verifiably done. A separate **checker agent** decides completion by actually reading the files and running the build or tests, not by trusting the worker's word.

## Autonomy levels

* **Supervised**: you approve actions as usual.
* **Safe-auto**: safe actions run automatically, risky ones ask.
* **Full-auto**: everything runs, with a live action log of every command it ran or auto-approved.

## Guardrails

* Token budget for the whole session.
* Stall detector that stops loops going nowhere.
* Stop-on-error option.
* Optional per-turn step cap.

## Isolation

Tick "Isolate in a git worktree" and the entire autonomous session works on one throwaway copy of your repo. Every turn builds on the last, and when the loop stops you review everything as a single set in **Proposed changes**: merge it or discard it.

{% hint style="info" %}
Even in full-auto, the git safety floor holds: no pushes, no history rewrites, no destructive git.
{% endhint %}