For teams accountable for what AI decides.

AI acts.Keep the receipt.

Policy before the model call. One signed record per decision. Verifiable without us.

Sample

Checking

Checking this receipt in your browser

Signed record

Checked in this browser. No call to Aqta.

Open verifier on npmChange one field above. The signature breaks.

Works with the models you already use.

Most tools audit after the fact. AqtaCore runs before the action.

When audit asks what the AI decided, you already have the signed receipt.

signing historyProduction gateway live

Signature schemeEd25519, canonical JSON

from runtime to independent proof

Policy runs before the model call. A reviewer checks the same receipt later with the published key. No callback to Aqta.

At the decision point

Your systemrequest

AqtaCorepolicy checked

Any modelonly if allowed

Blocked? The model is not called. The receipt remains.

Sample receiptATTESTATION-v1

Issued at runtime. Checkable later.

Outside Aqta

Independent check

A reviewer checks the same receipt with the published key.

For your engineering team

Get the receipt. Pin the policy. Drop in the proxy.

01
Sign every AI call
An Ed25519-signed receipt per call. Verifiable offline. No Aqta in the trust path.
- Open verifier on npm
- Verified offline, no callbacks
02
Pin the policy
Set per-user budgets, declare PII and PHI rules, detect loops. Every receipt records which policy version ran.
- Policies versioned by Git SHA
- Exported in audit bundles
03
Drop in the proxy
Point your OpenAI-compatible client at AqtaCore. One endpoint for any model.
- Any OpenAI-compatible SDK works
- Zero application rewrites