Docs / Trust

Safety and revert

About 3 minutes

Autopilot is only useful if you can leave it running. That comes from making every change it makes visible and reversible.

The activity journal

Every autonomous write is recorded in an activity journal you can open at any time. Each entry shows what changed, on which note, when, and a plain summary such as "Set type, tags, summary" or "Extracted 2 decisions, 2 open loops".

One-click revert

Before any automatic write, LocalBrain snapshots the previous version of the file. Revert restores that snapshot instantly. If a change looks wrong months later, it is still one click to undo, and reverting a change never loses your own edits.

Confidence you can read

Each change carries a confidence level of low, medium, or high. That level is grounded in the note itself, how much content it has and how many fields actually changed, not the model's opinion of its own work. A thin note gets low confidence because there was little to go on.

Fail closed. Anything destructive requires care by default, and you can put files or whole folders off limits with protected zones. Nothing happens that you cannot see and undo.
Open LocalBrain Back to Docs