A local AI agent that shows its work.

Try the trust loop without running a backend: pick a recipe, search local evidence, preview a diff, apply it, and inspect the receipt.

96Trust score

Search to Diff Preview

workspace/welcome.md Matched local evidence about receipts, explicit Apply, and diff-safe writes.

--- a/workspace/welcome.md
+++ b/workspace/welcome.md
@@
 AgentTrail runs locally with Ollama or an OpenAI-compatible local backend.
+Every search, preview, and apply action is saved as an auditable receipt.
+Writes stay off by default until you approve the exact diff.

Diff preview ready. Nothing has been written.