Declare it. Verify it. Sign it.
Three steps from an agent you hope behaves to an agent you can prove behaved.
Declare it. Verify it. Sign it.
Three steps from an agent you hope behaves to an agent you can prove behaved.
- 01
Declare
An Alignment Card states what the agent may and may not do — declared intent, permitted tools, data boundaries, escalation contract, drift budget — in plain, versioned, signed terms. Public at /.well-known/alignment-card.json.
- 02
Verify
Every decision is checked in flight against that card. Drift, policy breaks and risky reasoning are caught as they happen — verdict clear, review, or boundary violation — not in a post-mortem.
- 03
Sign
The result is a tamper-evident, Ed25519-signed record and — once an agent has 50 verified checkpoints — a portable Trust Rating from 0 to 1000 that anyone can verify independently, offline, without trusting Mnemom.
Catch it before it happens, not after.
Most AI governance tools tell you what happened. Mnemom governs what can happen — policy is evaluated before the action executes, and every evaluation leaves a cryptographic receipt.
Find your starting point
Read the docs, wire the gateway
Install, alignment card schema, checkpoint coverage, OpenTelemetry export.
I'm evaluating the riskInspect the proof boundaries
Threat model, what a signature does and does not prove, compliance mapping.
I'm planning the rolloutWork out what a fleet costs
Everything is included; the only variable is the μ you use. Simulate it on the pricing page.
Know what your agents are actually doing. Before the board asks.
Two commands to install. A signed record from the first decision onward — or talk to us about a self-hosted, dedicated-cell deployment.
