What is Adjudon?
Adjudon is a runtime compliance layer for AI agents — trace every decision, enforce policies, and prove EU AI Act compliance.
What it does
- Traces: Every AI decision is recorded with input, output, confidence score, and audit trail
- Policies: Define rules that automatically approve, flag, or block AI decisions
- Human Review: Flagged decisions route to a review queue for human oversight
How it works
Your AI Agent → Adjudon SDK → POST /api/v1/traces → Policy Engine → Approved/Flagged/Blocked
↓
Review Queue (if flagged)
↓
Audit Log (SHA-256 chain)
EU AI Act coverage
| Article | Status | What it covers |
|---|---|---|
| Art. 13 — Transparency | Live | Every AI decision traced with input, output, confidence, audit trail |
| Art. 14 — Human Oversight | Live | Review queue routes flagged decisions to human reviewers |
| Art. 12 — Record-Keeping | Roadmap Q3 2026 | Dedicated logging features |
| Art. 26 — Deployer Obligations | Roadmap Q3 2026 | Compliance packs for deployers |
Architecture principles
Fail-open: If Adjudon is unreachable, your agent continues. The SDK retries in the background and returns passthrough status — your agent is never blocked by compliance infrastructure.
EU-only data: All data is stored in Frankfurt (eu-central-1). No data leaves the EU.
Under 25ms p95: Trace ingestion adds less than 25ms at the 95th percentile. Fast enough to be invisible.