How it works
From alignment card to cryptographic proof in four steps.
No trust required. Every step produces independently verifiable evidence. Open protocols at every layer.
Declare alignment with a signed card
Publish an Alignment Card under the Agent Alignment Protocol (AAP): declared intent, permitted tools, data boundaries, escalation contract, drift budget. The card is your agent's contract — public at /.well-known/alignment-card.json, machine-readable, audit-anchored.
Safe House screens every transaction at four checkpoints
Every inbound prompt, agent action, and outbound payload is evaluated at four checkpoints — front door, back door, inside.autonomy, inside.integrity — each independently set to off, observe, nudge, or enforce. AAP supplies the declared boundary. AIP supplies the in-flight integrity verdict. Mnemom AEGIS adds cross-tenant signed Managed Rules on top — sub-30s P95 propagation, Ed25519-signed, with tier-1 and tier-2 rules under dual-control human review.
Verify cryptographic proofs (AIP + CLPI)
AIP produces an IntegrityCheckpoint with verdict clear, review-needed, or boundary-violation — every thinking block SHA-256 hashed for tamper evidence. CLPI governs the card lifecycle across five phases, anchors Trust Ratings on Base L2 via ERC-8004, and emits OpenTelemetry traces. Anyone can verify — no trust required.
Monitor Trust Ratings and reputation
Scores refresh every 6 hours from live protocol data across five dimensions: integrity ratio, compliance, drift stability, trace completeness, and coherence. A single number from 0 to 1000, with weekly snapshots for trend history — and a Coherence framework that scores multi-agent fleet trust.
See it in action.
Watch cryptographic verification run live in your browser, or install the SDK and start verifying in under 5 minutes.
