Skip to content

Mohar documentation

Mohar is an open-source trust layer for AI agents. It records every agent action as a signed, hash-chained receipt, lets an agent earn permission one capability at a time from outcomes that were actually checked, takes that permission away the moment a gated action fails, and sends only the actions an agent hasn't earned to a person.

Status

Mohar is early. The receipt log and the mohar CLI are released (0.1). The trust engine, the escalation flow and the host adapters are designed and in progress; the roadmap and its milestones say exactly what exists.

Start here

  • Install and first log: install mohar, record decisions, verify the chain, and see what happens when someone edits or truncates it.
  • CLI reference: every command and flag.
  • Design: capabilities, trust levels and the rules that move them, verification, escalation, receipts, privacy and the threat model.
  • Receipts and checkpoints: the formats, with test vectors and an independent checker.
  • Integrations: how Mohar plugs into Claude Code, Codex, CAO, ADK, MCP and A2A, and how strong each seam is.
  • Compliance: how Mohar maps onto HIPAA, GDPR, SOC 2, ISO 27001 and 42001 and the EU AI Act, control by control.

Versions

These docs are versioned. The selector at the top switches between latest (the newest release), each release, and dev (the main branch). Docs for a release never change after it ships.