Documentation

gmonk documentation

gmonk is a self-hosted safety and disaster-recovery layer for AI agents. It records every action an agent takes, blocks the ones it can't undo, and lets you reverse a run — drift-aware — when one goes wrong. This is the complete reference.

The mental model

gmonk decides on physics, not policy. Instead of asking "is this action allowed?" it asks "can this action be undone?" — a property of the action itself. That one question drives both halves of the product:

• Prevention — before an action runs, the reversibility oracle scores it and the gate refuses anything unrecoverable.
• Recovery — every change is captured with its before-state, so a human can reverse a run per-action, drift-aware, after the fact.

They're the same engine pointed in opposite directions in time. Start with what gmonk covers, then read capture, prevention, and recovery.

All docs

Start here

How it works

Operate