Commercial tiers now in private beta with design partners · Book an agent audit →

When your agents act on their own, can you prove what they did?

AIR signs every agent action, binds it to the human who authorized it, and anchors it to a public log. One loop, four moves.

AUDIT | PROVE | PROTECT | MONITOR

Start free Book a demo →

Open source · MIT · no credit card · verify it yourself

212 agents live

Ed25519 signed

BLAKE3 chained

· Global agent network · live ·

02:14:07 · agent: records-export-07

An agent with database access just exported 9,000 patient records.

Your team can tell you it happened. They can't prove what it was authorized to do, or that the log wasn't edited after. That's the question every regulator, auditor, and court is about to ask.

Audit it. Prove it. Protect it. Monitor it.

One loop, running the whole time, so the answer exists before you need it.

Catch what breaks scope, the instant it happens.

16 detectors read every action in real time. The deterministic floor halts an agent the moment it steps outside what a human authorized.

See detection →

Signed Intent Capsule · chain

Delegation · dr.okafor authorized claims-bot

Action · ehr.read — in scope

Halt · bulk export blocked — SV-EXFIL

✓ chain intact · verify on search.sigstore.dev

Prove it to anyone, without their trust.

Every action is signed in-process, bound to the authorizing human, and anchored to a public log. Self-authenticating evidence an auditor, insurer, or court can verify.

See the evidence →

Containment

Agent halted · awaiting approvalHeld

Approve & sign · Auth0 / passkeyFIDO2

No agent acts without a delegation. Forged tokens stay halted.

Turn AIR on.

And from this moment, every action is on the record.

This page proves it.

Every request you just made was signed and anchored to a public transparency log. Not a claim, a dare. Verify it yourself, with zero Vindicara infrastructure in the path.

› verify on search.sigstore.dev

OWASP AgenticEU AI Act Art. 72NIST AI RMFFRE 902(13)–(14)MIT open source