Every agent action, hash-chained and human-decided.
The Loom is the governance layer the 2026 standards demand — a hash-chained, append-only audit trail where Lumi proposes and a human decides. Post-quantum identity, independent verification, co-presence enforced at the gate.
What "loom" actually means
Try it. Tamper with block 4 and watch the chain break — then verify.
This is a simplified demo of the property. The real Loom chain uses SHA-256 over RFC 8785-canonicalized records, and the live chain is independently verifiable.
What's inside
Hash-chained audit
SHA-256 per RFC 8785, tamper-evident, independently verifiable via GET /api/loom/verify.
Co-presence enforcement
Lumi proposes, the human decides — a scope gate refuses any autonomous action.
Post-quantum identity
ML-DSA-65 capability tokens bound to a responsible principal (AIC-JWT).
Refusal sidecar
Every refused action recorded with its reason, on a separate verifiable chain.
Built with
Questions people ask
Is the chain really verifiable?
Yes — GET /api/loom/verify recomputes it. Anyone can check it.
What stops Lumi from acting alone?
A scope gate refuses any action the human did not approve.