From shadow AI to verifiable evidence — in one afternoon.
An AiGovOps Foundation project, implementing the OVERT 1.0 open standard.
Standard steward: Glacis Technologies · Foundation Founding Steward Partner.
Beacon is an OVERT 1.0–conformant runtime that produces signed, verifiable receipts of every AI decision. It runs alone.
-01696F?style=flat-square){kind=tracking}
Beacon implements the OVERT 1.0 open standard for observable verification evidence at the AI runtime boundary, stewarded by Glacis Technologies, Inc. Beacon is being registered as an OVERT Protocol Profile under the royalty-free patent covenant published at overt.is/ipr-policy.
| Standard | OVERT 1.0 (Glacis Technologies, 25 March 2026) |
| Profile | aigovops-beacon.v1 — registration in progress, see docs/PROFILE_REGISTRATION.md |
| Target conformance | AAL-1 (self-declared) at registration · AAL-2 with IAP on roadmap |
| Patent covenant | Royalty-free under overt.is/ipr-policy |
| Crosswalk | crosswalks/overt-mapping.yaml — Beacon controls → OVERT six domains |
The AiGovOps Foundation does not issue standards. We implement, adopt, teach, and build community around them. See STANDARDS.md, GOVERNANCE.md, and STEWARD.md.
The project's home page is a self-running demo: it generates an Ed25519 key, discovers a sample inventory, signs each receipt with canonical JCS (RFC 8785), and assembles a real downloadable evidence bundle — all client-side, no server.
Live site (GitHub Pages): https://aigovops-foundation.github.io/aigovops-beacon/
The page also hosts:
- The full audit framework PDF (NIST AI RMF + EU AI Act Art. 13 + ISO/IEC 42001 + HIPAA + Human Flourishing)
- The five individual checklist PDFs and the Excel inventory template
- The starter .zip for the entire repo
- One-click deploy recipes for Railway, Render, and DigitalOcean App Platform
Want to read the home-page source? It's plain HTML + CSS + JS in docs/.
Most AI governance today is a PDF and a meeting. That's not governance — that's aspiration.
Beacon is the simplest possible answer to one question your auditor actually asks:
"What AI models are running on our network — who is using them, with what version, against what data, and where is the evidence?"
It answers that question in two ways at the same time:
- Beacon Studio — a five-step, sticky-notes-friendly wizard. A governance lead with zero terminal experience walks in cold and walks out with a signed checklist, an evidence trail, and a pull request that turns the checklist into policy-as-code.
- Beacon Control Plane — the same data, exposed as a power-user dashboard. Filters, raw event stream, OPA/Rego console, SIEM export. Same backend. Same evidence. Toggle in the top right.
One engine. Two front doors. Zero PDF theater.
| Layer | What it does |
|---|---|
| Discovery | Detects AI usage across your network via passive DNS/HTTP scan, outbound proxy hooks, cloud account scans (AWS Bedrock, Azure OpenAI, Vertex, Databricks), CSV import, and an optional endpoint agent. |
| Inventory | Every detection becomes a row: vendor · model · version · occurrence count · users · first seen · last seen · environment · evidence pointer. |
| Transaction tracking | Every observed interaction becomes a Replay-style receipt: logged-in user · UTC timestamp · model · version · prompt (or hash) · result (or hash) · Ed25519 signature. Append-only NDJSON. Merkle root anchored hourly. |
| Checklists | Pre-built packs: NIST AI RMF, EU AI Act Article 13, ISO/IEC 42001, HIPAA-for-AI, and the Foundation's Human Flourishing Gate. Auditor checks what applies. |
| Policy as Code | One button emits a gate.<org>.yaml + OPA .rego + Governance Decision Record + signed bundle. Opens a PR against your governance repo. |
| Two front doors | /studio for the workshop, /control for the power user. |
Corrected 2026-07-20. Both commands here were wrong. The Docker option pulled
ghcr.io/bobrapp/aigovops-beacon:latest, which is not pullable — the repo moved to theaigovops-foundationorg and no public image was ever pushed. The local option rannpm installat the repo root, where there is nopackage.json; the Node project lives inserver/. A quickstart that fails on its first command is worse than no quickstart.
# Local — this is the path that works today
git clone https://github.com/aigovops-foundation/aigovops-beacon
cd aigovops-beacon/server
npm install
npm run init # first run only: creates the local store
npm run dev # http://localhost:8787(The old text said port 8080. The default in server/src/lib/config.js is 8787 — so even
the port in the broken quickstart was wrong. Verified by reading the config, not by assuming.)
A published container image is not available yet. When one is, it will be pushed to
ghcr.io/aigovops-foundation/aigovops-beacon and documented here — not before.
Open http://localhost:8787. Studio opens by default. Click the toggle in the top right for the Control Plane.
Four more one-click deploys live in /deploy: Railway, Render, DigitalOcean App Platform, and Docker Compose.
The point of all of this is the last step: someone who has never heard of Beacon confirms your evidence is what you say it is. Every exported bundle carries its own verifier, so that is one command with nothing installed:
python3 verify_bundle.py .beacon-verify: OK — 47 receipts verified (bundle format)
manifest: intact (sha256 df9708e7aa774938…, over file)
keys: 2 (ab12cd34ef561234, 9f81be22c7a40d55)
receipts: 2026-07-29.ndjson (12)
receipts: 2026-07-30.ndjson (35)
It checks that the manifest is intact, that every receipt's Ed25519 signature verifies over its RFC 8785 canonical bytes against the key that signed that receipt — bundles spanning a key rotation included — and that no receipt was dropped. Exit 0 means clean; non-zero names the failures.
Requirements: Python 3.10 or newer. Nothing else. cryptography is used when
it happens to be installed; when it is not, a pure-Python Ed25519 verifier in
the same file takes over, so an air-gapped review machine works too. No
network calls, ever — and the file is short enough to read before running it.
verify_bundle.py is a copy of src/beacon_verify.py.
The same tool verifies a single receipt log or the chained audit log:
python3 src/beacon_verify.py --public-key ~/.beacon/keys/ed25519.pub \
~/.beacon/receipts/2026-07-30.ndjsonThis is what a non-technical auditor sees. No terminal. No YAML. No jargon.
| Step | Title | Auditor sees |
|---|---|---|
| 1 | What network are we looking at? | "Paste a domain, upload a CSV, or connect a cloud account. Beacon will look for AI models running there." |
| 2 | Beacon is looking… | Live tile board of every model detected. Vendor logo, name, version, count, last-seen, environment. Updates in real time. |
| 3 | Pick what matters. | Tap rows that are in scope. Beacon explains each in plain English ("This is OpenAI's GPT-4o, used 1,840 times this week by 12 people in your support team."). |
| 4 | Pick your guardrails. | Cards for NIST RMF, EU AI Act, ISO 42001, HIPAA-for-AI, Human Flourishing. Auditor checks the ones that apply. |
| 5 | Here's your audit. | One screen: traffic-light checklist, the receipts behind each green check, and a Make it Policy as Code button. |
Full spec, copy, and accessibility notes in docs/STUDIO_FLOW.md. Workshop facilitator guide in docs/AUDITOR_WORKSHOP.md.
This repo dogfoods the Foundation's operating thesis:
Agents do the bureaucracy; humans hold the meaning — and humans hold the keys.
Beacon will draft anything — the inventory, the checklist mapping, the policy YAML, the GDR, the PR description. But every human-decision gate stays a human-decision gate: scope ratification, exception grants, trust-tier promotion, and policy publication all require a named human approver. The receipt records who approved what, when, and with which key.
| # | Pillar | How Beacon delivers it |
|---|---|---|
| 01 | Governance as Code | Every checklist exports to versioned YAML + Rego. The Studio's final button is literally git push. |
| 02 | AI Technical Debt Elimination | Discovery surfaces every model your org didn't know was running. Un-receipted activity is technical debt — visible, countable, closeable. |
| 03 | Operational Compliance | Receipts are Ed25519-signed and Merkle-anchored. Compliance becomes a query, not a quarterly scramble. |
| 04 | Community-Driven Standards | Checklist packs are plain YAML. Fork, edit, PR. The community owns the controls. |
Beacon doesn't replace anything — it sits upstream of the other Foundation projects and feeds them:
┌───────────────────────────┐
│ AiGovOps Beacon │ ← you are here
│ (discovery + studio) │
└────────────┬──────────────┘
│ signed receipts + checklist bundle
▼
┌───────────────────────────┐ ┌─────────────────────────┐
│ aigovops-Replay │ │ aigovops-foundation-os │
│ (receipt verification) │ │ (policy + GDR schemas) │
└────────────┬──────────────┘ └────────────┬────────────┘
│ │
▼ ▼
┌───────────────────────────┐ ┌─────────────────────────┐
│ webhook-sentinel │ │ aigovops-prompt-studio │
│ (PR-review trust) │ │ (prompt-level audit) │
└───────────────────────────┘ └─────────────────────────┘
Receipt schema is wire-compatible with aigovops-Replay. Gate schema is wire-compatible with aigovops-foundation-os.
For the broader Foundation context — sister projects, the umbrella policy-as-code companion, and how Beacon relates to other compliance tooling — see RELATED.md. Beacon does not require any of those projects.
Beacon is fast enough that you'll never be tempted to disable it under load. On commodity x86_64 hardware:
- 9,000+ receipts/sec signed by a single process
- 84 µs p50 inline overhead per AI call (canonicalize + sign)
- 8,000+ verifies/sec for stream verification
Full methodology, capacity-planning numbers, and a reproducible benchmark script in BENCHMARKS.md.
Beacon ships a full test pyramid across both runtimes — the Node/Express server and the Python beacons/SDK. Tests are deterministic: anything random or fuzz-driven is seeded, with an environment-variable override for local exploration.
| Layer | What it checks | Node | Python |
|---|---|---|---|
| Unit | Pure functions: canonicalization, signing/verify, receipt building | npm test (in server/) |
pytest tests/unit |
| E2E | Live routes answer and responses conform to the receipt schema | npm run test:e2e |
pytest tests/e2e |
| Chaos | Fuzzed/malformed input + injected I/O faults never crash the server | npm run test:chaos |
pytest tests/chaos |
| Scale | 10k+ receipts signed/bundled within a time & memory budget (gated) | — | pytest -m scale tests/scale |
# Node server tests (unit + E2E + chaos via auto-discovery)
cd server && npm install && npm test
# Python tests (unit + E2E + chaos; scale is excluded by default)
pip install -e ".[dev]"
python3 -m pytest # everything except scale
python3 -m pytest -m scale tests/scale # the heavy scale layer, on demandDeterminism knobs: CHAOS_SEED / CHAOS_RUNS (Node), HYPOTHESIS_SEED /
CHAOS_MAX_EXAMPLES (Python), and SCALE_COUNT / SCALE_TIME_BUDGET /
SCALE_MEM_BUDGET / SCALE_SEED (scale). See
CONTRIBUTING.md for what each layer expects from a new PR.
The complete HTTP API is documented under docs/api/:
openapi.yaml— OpenAPI 3.1 spec for every endpoint: methods, paths, request/response schemas (success and error), auth, and idempotency notes.flows.md— how a receipt travels the system: creation → signing → bundle assembly → anchoring → consumption by Lantern, with Mermaid sequence diagrams.actions.md— every action verb (gate.evaluated,bundle.signed,admission.allowed,inference.observed, …) with semantics, required fields, and examples.
aigovops-beacon/
├── README.md ← you are here
├── LICENSE ← Apache-2.0
├── CONTRIBUTING.md
├── CODE_OF_CONDUCT.md
│
├── docs/
│ ├── STUDIO_FLOW.md ← five-step wizard spec
│ ├── CONTROL_PLANE.md ← advanced-mode reference
│ ├── ARCHITECTURE.md ← discovery → receipts → policy pipeline
│ ├── RECEIPT_SCHEMA.md ← Ed25519 transaction-record spec
│ └── AUDITOR_WORKSHOP.md ← 90-minute facilitator guide
│
├── checklists/ ← starter audit packs (YAML)
│ ├── nist-ai-rmf.v1.yaml
│ ├── eu-ai-act-art13.v1.yaml
│ ├── iso-iec-42001.v1.yaml
│ └── hipaa-for-ai.v1.yaml
│
├── policy/ ← outputs the Studio emits
│ ├── gate.beacon.v1.yaml
│ ├── gate.human-flourishing.v1.yaml
│ └── rego/
│ ├── production_readiness.rego
│ └── transaction_signing.rego
│
├── server/ ← Node/Express engine
│ ├── package.json
│ └── src/
│ ├── index.js
│ ├── routes/ ← /discover /receipts /checklist /export
│ ├── services/ ← discovery, signing, checklist
│ └── lib/ ← Ed25519, merkle, ndjson
│
├── studio/ ← React/Vite five-step wizard
│ ├── package.json
│ └── src/
│ ├── App.jsx
│ ├── pages/ ← Step01..Step05, Control
│ └── components/
│
├── deploy/
│ ├── Dockerfile
│ ├── docker-compose.yml
│ ├── railway.json
│ ├── render.yaml
│ └── do-app.yaml
│
└── scripts/
├── verify-receipt.js
└── seed-demo.js
- License: Apache-2.0 — same as
aigovops-foundation-os. - Code of Conduct: Contributor Covenant v2.1.
- Community: AiGovOps Foundation. Founders: Ken Johnston and Bob Rapp.
"Governance that ships with the code, not after." — Bob Rapp
"Shadow AI to verifiable evidence." — Ken Johnston
Tagline: YES-Ship AI · YES-Steady AI · YES-Recover AI
- Bob Rapp — bob.rapp@aigovops.community
- Ken Johnston — ken.johnston@aigovops.community
- Foundation — aigovopsfoundation.org
Verifiable AI governance — Apache-2.0, no SaaS lock-in.
- Redwood v2 (draft FEP) — Foundation Enhancement Proposal currently in WG bootstrap. Tracks ratification of receipt schemas, UCID registry, and viability lens (Ashby + Beer + sociotechnical). See the ratification project and v0.1.0-draft release.