Traceable work for humans and agents.

Intelligence that moves across time, devices, and contexts. Mentu keeps work traceable with a simple protocol and a commitment ledger, so every promise and closure can be reconstructed.

A commitment ledger is an append-only record that links observations to commitments, and commitments to evidence, so accountability is reconstructable.

• Carry context: move across devices and threads without losing continuity
• Make intent explicit: commitments are named, owned, and reviewable
• Close with proof: done means evidence, not a checkbox

Observation (memory) → Commitment → Evidence → Closure

Mentu is designed around two invariants:

• Append-only ledger: nothing is edited, nothing is deleted
• State by replay: current state is computed from the ledger, not stored elsewhere

A typical workspace contains:

• .mentu/ledger.jsonl (the ledger)
• genesis.key (optional workspace constitution for identity, permissions, and validation tiers)

If you want to try the protocol locally:

npm install -g mentu mentu init

mentu capture "Customer reported checkout bug" mentu commit "Fix checkout bug" --source <memory_id> mentu claim <commitment_id>

## do the work...

mentu capture "Fixed null check in payment.ts:42" --kind evidence mentu close <commitment_id> --evidence <evidence_memory_id>

•	mentu: the protocol specification and core design
•	mentu-cli: the official CLI implementation

•	Evidence-based closure: closures require proof, not a flag
•	Review airlock: optional in_review flow for agent workflows and higher-risk work
•	Tiered validation: optional policies in genesis.key to gate what can auto-close
•	Multi-agent accountability: patterns like the Dual Triad help separate creation from verification

Issues and PRs are welcome. If you are proposing a new operation, state transition, or validator, include: • the intent • expected ledger operations • replay semantics • failure modes and auditability

See each repository for licensing details.