Content
# Sovereign Matrix
> **Audit-grade AI agents. Every output cryptographically signed. Apache-2.0 verifiers in three languages.**
Six-framework regulatory exporter suite + the receipt primitive
underneath. Receipts in, regulator-ready reports out. Closed-source
vendors charge $5K-300K/year per framework; we ship all six as
Apache 2.0.
[](./src/__tests__)
[](./LICENSE)
[](./LICENSE)
[](https://sovereignmatrix.agency)
[](./packages/mcp)
[](./packages/verifiable-receipts)
## The 12-package stack
| Package | Purpose | Status |
| ------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------ |
| [`@sovereign-matrix/verifiable-receipts`](./packages/verifiable-receipts) | Core: Ed25519 + ML-DSA-65 (FIPS 204) post-quantum dual-signing, RFC 9162 transparency log, 42 Guardian rule packs, 3-language verifier (TS / Python / Go) with byte-deterministic conformance corpus | v0.3.0 |
| [`@sovereign-matrix/openai-receipts`](./packages/openai-receipts) | OpenAI SDK wrapper — 3 lines to mint signed receipts around chat completions | v0.1.0 |
| [`@sovereign-matrix/anthropic-receipts`](./packages/anthropic-receipts) | Anthropic SDK wrapper | v0.1.0 |
| [`@sovereign-matrix/google-receipts`](./packages/google-receipts) | Google Gemini SDK wrapper | v0.1.0 |
| [`@sovereign-matrix/ai-sdk-receipts`](./packages/ai-sdk-receipts) | Vercel AI SDK universal wrapper (OpenAI / Anthropic / Google / Mistral / Cohere) | v0.1.0 |
| [`@sovereign-matrix/annex-iv`](./packages/annex-iv) | **EU AI Act Article 11 + Annex IV** technical-documentation exporter | v0.1.0 |
| [`@sovereign-matrix/iso-42001`](./packages/iso-42001) | **ISO/IEC 42001:2023 AIMS** — clauses 4-10 + 38-control Annex A | v0.1.0 |
| [`@sovereign-matrix/nist-ai-rmf`](./packages/nist-ai-rmf) | **NIST AI RMF 1.0** — GOVERN / MAP / MEASURE / MANAGE profile | v0.1.0 |
| [`@sovereign-matrix/soc2-evidence`](./packages/soc2-evidence) | **SOC 2 AICPA Trust Service Criteria 2017** evidence binder | v0.1.0 |
| [`@sovereign-matrix/gdpr-dpia`](./packages/gdpr-dpia) | **GDPR Article 35 DPIA + Article 30 RoPA** | v0.1.0 |
| [`@sovereign-matrix/hipaa-security`](./packages/hipaa-security) | **HIPAA Security Rule** (45 CFR § 164.308-318) | v0.1.0 |
| [`@sovereign-matrix/mcp`](./packages/mcp) | MCP server exposing all of the above as tools to Claude Code / Cursor / Zed | v0.1.0 |
## Three lines to start
```ts
import Anthropic from "@anthropic-ai/sdk";
import { mintMessageReceipt } from "@sovereign-matrix/anthropic-receipts";
const client = new Anthropic();
const message = await client.messages.create({
model: "claude-sonnet-4-6",
max_tokens: 1024,
messages: [{ role: "user", content: "Summarize this contract..." }],
});
// ← This is the only added line:
const receipt = await mintMessageReceipt(message, { sign, agentSlug, runId });
// receipt.signature → "v2=<base64>" — verifiable forever
// receipt.overall → "pass"
```
Same shape for [`openai-receipts`](./packages/openai-receipts),
[`google-receipts`](./packages/google-receipts), and
[`ai-sdk-receipts`](./packages/ai-sdk-receipts) (universal, works with
every Vercel AI SDK provider).
## Why this exists
Every regulated AI buyer asks the same question:
> "If the model makes a decision the regulator later challenges, what
> evidence do we hand the court that it was defensible at the moment
> it was made?"
The vendor answer is usually "we logged it somewhere." This is the
**open-source answer that travels with the call** — a
cryptographically-signed, post-quantum-secure receipt any third party
can verify with the issuer's public key.
Then layer on the regulatory exporters: feed your receipt set into
any of the six framework packages and emit an auditor-ready binder
in Markdown + JSON. Byte-deterministic. The auditor with your
receipts can reproduce the report exactly.
## Six frameworks, one receipt set
| Framework | Closed-source competitor | Their typical price | Sovereign Matrix |
| ------------------- | ---------------------------------------- | ------------------- | ---------------- |
| EU AI Act Annex IV | Credo AI, Holistic AI, IBM watsonx | $50-200K/yr | Apache 2.0 |
| ISO/IEC 42001:2023 | Holistic AI + cert bodies (BSI, TÜV SÜD) | $50-150K/yr + cert | Apache 2.0 |
| NIST AI RMF 1.0 | IBM watsonx.governance, Credo AI | $100-300K/yr | Apache 2.0 |
| SOC 2 evidence | Vanta, Drata, Secureframe | $5-50K/yr | Apache 2.0 |
| GDPR DPIA + RoPA | OneTrust, TrustArc, BigID | $10-100K/yr | Apache 2.0 |
| HIPAA Security Rule | HITRUST, Compliancy Group | $20-100K/yr + cert | Apache 2.0 |
Single receipt set + six exporters + one function call per framework.
Tag the receipt once with the right Guardian pack; every framework's
report counts it as evidence.
## MCP — talk to your compliance pipeline from Claude Code
```json
// claude_desktop_config.json
{
"mcpServers": {
"sovereign-matrix": {
"command": "npx",
"args": ["-y", "@sovereign-matrix/mcp"]
}
}
}
```
Restart Claude → 7 new tools appear:
```
verify_receipt · build_annex_iv · build_iso_42001 · build_nist_ai_rmf ·
build_soc2_evidence · build_gdpr_dpia · build_hipaa_security
```
Ask Claude: _"Build an EU AI Act Annex IV report from
receipts/q2-2026.json with this system metadata: …"_ → Claude calls
the deterministic exporter against your real receipts and writes
auditor-ready Markdown back to your repo.
## Live preview
- **[/compliance](https://sovereignmatrix.agency/compliance)** — hub linking all 6 exporters
- **[/compliance/annex-iv](https://sovereignmatrix.agency/compliance/annex-iv)** — EU AI Act preview
- **[/compliance/iso-42001](https://sovereignmatrix.agency/compliance/iso-42001)** — AIMS preview
- **[/compliance/nist-ai-rmf](https://sovereignmatrix.agency/compliance/nist-ai-rmf)** — US federal preview
- **[/compliance/soc2](https://sovereignmatrix.agency/compliance/soc2)** — Trust Service Criteria binder
- **[/compliance/gdpr-dpia](https://sovereignmatrix.agency/compliance/gdpr-dpia)** — DPIA + RoPA
- **[/compliance/hipaa](https://sovereignmatrix.agency/compliance/hipaa)** — Security Rule binder
Every preview is server-rendered against a sample receipt set so you
see the exact bytes the exporter produces before you install anything.
## Three-language verifier
The core receipt primitive ships verifiers in TypeScript, Python, and
Go. Every verifier runs the same conformance corpus and MUST produce
byte-identical pass/fail outcomes:
```
packages/verifiable-receipts/conformance/
├── fixtures/ # 10 JSON receipt fixtures
│ ├── basic-pass.json
│ ├── inclusion-7leaf-idx6.json # the carry-up edge case
│ ├── threshold-3of5.json
│ └── ...
├── verifier.ts # TypeScript verifier
├── verifier.py # Python verifier
└── verifier.go # Go verifier
```
Auditor clones the repo, runs three commands, confirms byte-identical
outcomes across all three languages. Same discipline as Sigstore +
Certificate Transparency + JOSE conformance suites.
## Architecture
```
┌──────────────────────────────────────────────────────────────────┐
│ Your AI call (OpenAI / Anthropic / Google / Vercel AI SDK) │
└────────────────────────────┬─────────────────────────────────────┘
│ + 3 lines of wrapper code
▼
┌──────────────────────────────────────────────────────────────────┐
│ @sovereign-matrix/*-receipts wrapper │
│ → Mints a VAOS receipt (Ed25519 + ML-DSA-65 dual-signed) │
│ → Optionally runs Guardian rule packs (42 packs · 7 continents) │
│ → Anchors to RFC 9162 transparency log │
└────────────────────────────┬─────────────────────────────────────┘
▼
┌──────────────────────────────────────────────────────────────────┐
│ Receipt set (your audit trail — 90 days / 365 days / forever) │
└────────────────────────────┬─────────────────────────────────────┘
┌────────────┬───┴───┬─────────────┬──────────────┐
▼ ▼ ▼ ▼ ▼
annex-iv iso-42001 nist-rmf soc2-evidence gdpr-dpia/hipaa
(EU AI Act) (AIMS) (US fed) (TSC binder) (privacy/health)
│ │ │ │ │
└────────────┴───────┴─────────────┴──────────────┘
▼
Auditor-ready Markdown + GRC-tool JSON
Byte-deterministic. Reproducible by your regulator.
```
## Methodology
[`docs/methodology/SOVEREIGN-METHODOLOGY-V2.md`](./docs/methodology/SOVEREIGN-METHODOLOGY-V2.md)
is a CC0 deep-dive on the Opus 4.7 prompt patterns deployed across
the platform — CRISPE + XML + `<search_first>` anti-hallucination
contracts + multi-agent delegation patterns. 10 sections, field-tested
production prompts.
## Tests + gates
```
$ npm test
✓ 3852 tests passing (256 files)
✓ 4 skipped
$ npx tsc --noEmit
✓ clean
$ npm run lint
✓ 0 errors, 21 baseline warnings
$ npm run build
✓ 354 static pages
✓ 9 compliance routes prerendered
```
CI runs lint + typecheck + tests + security audit + SBOM generation
- dependency review on every PR.
## Repository layout
```
sovereign-v2/
├── packages/
│ ├── verifiable-receipts/ # Core: receipt primitive, transparency log, packs
│ ├── verifiable-receipts-py/ # Python verifier
│ ├── verifiable-receipts-go/ # Go verifier
│ ├── openai-receipts/ # OpenAI SDK wrapper
│ ├── anthropic-receipts/ # Anthropic SDK wrapper
│ ├── google-receipts/ # Google Gemini wrapper
│ ├── ai-sdk-receipts/ # Vercel AI SDK universal
│ ├── annex-iv/ # EU AI Act exporter
│ ├── iso-42001/ # ISO 42001 AIMS exporter
│ ├── nist-ai-rmf/ # NIST AI RMF profile
│ ├── soc2-evidence/ # SOC 2 evidence binder
│ ├── gdpr-dpia/ # GDPR DPIA + RoPA
│ ├── hipaa-security/ # HIPAA Security Rule
│ └── mcp/ # MCP server (Claude Code / Cursor)
├── src/ # Next.js 16 app (sovereignmatrix.agency)
│ ├── app/ # 354 static + ~50 dynamic routes
│ ├── lib/ # 273 lib modules
│ └── db/ # Drizzle ORM, 38 tables
├── docs/
│ ├── methodology/ # Sovereign Methodology v2 (CC0)
│ ├── partnerships/ # 5 sender-ready partnership briefs (CC0)
│ ├── launch/ # Show HN draft + launch checklist
│ └── specs/ # IETF Internet-Draft for VAOS
├── server/
│ ├── ws.ts # WebSocket server (real-time agents)
│ └── python-agents/ # Standalone Python agents
└── scripts/
├── publish-all.sh # One command publishes all 12 to npm
└── build-packages.mjs # Dependency-ordered workspace build
```
## Standards trajectory
- **IETF Internet-Draft**: `docs/specs/ietf-draft-vaos-00.md` —
formalises VAOS as a vendor-neutral wire format candidate for AI
agent operation receipts. Submission-ready for the IETF Independent
Stream.
- **CC0 spec**: anyone can implement a VAOS verifier without
permission, without royalty, without license risk.
- **Apache-2.0 reference verifiers**: TypeScript, Python, Go, with
byte-deterministic conformance.
## Contributing
See [CONTRIBUTING.md](./CONTRIBUTING.md). PRs welcome — every contribution
goes through the same CI gates as internal commits (lint + typecheck +
tests + security audit).
## Security
See [SECURITY.md](./SECURITY.md). 90-day responsible-disclosure window.
Email: `security@sovereignmatrix.agency`.
## License
| Layer | License |
| ------------------------- | ------------------------ |
| `packages/*` | Apache 2.0 |
| `docs/methodology/*` | CC0 (public domain) |
| `docs/partnerships/*` | CC0 (use freely) |
| `docs/specs/*` | IETF Trust per BCP 78/79 |
| Application code (`src/`) | MIT |
## What's next
- npm publishing: `./scripts/publish-all.sh` (12 packages → live on npm)
- Show HN: draft at `docs/launch/SHOW-HN.md` (Variant A — regulatory framing)
- IETF submission: `docs/specs/ietf-draft-vaos-00.md`
- Partnership outreach: 5 sender-ready briefs in `docs/partnerships/`
EU AI Act Article 6 enforcement: **2026-08-02**.
