VIGIL
VIGIL is the independent behavioral ledger and trust gate for production AI agents. It watches what agents do to each other and to their own controls, scores it as a Behavioral Integrity Index, holds outputs that fail the gate through an authority the agents cannot reach, and keeps a hash-chained record an auditor can verify without trusting us.
The promise. Every agent decision in your fleet is scored, gated, and sealed before it propagates, by a system your agents cannot see, edit, or shut off.
What it is not
Not a prompt-injection firewall. Not an observability tracer. Not an agent builder. Agent Studio builds. VIGIL watches, gates, and proves.
How it fits
| Layer | What it does |
|---|---|
| Detection | Scores every incoming event over per-type windows; four signals produce one BII |
| Enforcement | Evaluates the gate (pass, hold, alert, collapse) and records an enforcement action with a review flag |
| Audit | Appends a hash-chained, HMAC-sealed ledger record per event; the database rejects UPDATE and DELETE |
Your agents send events through an SDK, an adapter, or OpenTelemetry. The API answers synchronously with a gate decision. Reviewers decide holds through a separate service with its own database role. Nothing in the scoring path can release a hold.
Where to start
- Quickstart: a key, three lines of code, your first gate response.
- Concepts: what the score means and what each gate obliges you to do.
- API reference: every route, its scope, and its shapes, generated from the running service.
Check the service
The API is at https://vigil.supertruth.ai. /health needs no key and runs SELECT 1 against the database.
curl -sf "$VIGIL_URL/health"
Expected: {"status":"ok","service":"vigil","db":"ok"}.
Questions: louis@supertruth.ai.