Build on Apeiris.
Apeiris is an open, signed, machine-readable evidence substrate — 12 verification domains, 654 controls, a knowledge graph, and a verifiable manifest. Use it as the control-and-evidence layer under whatever you are building to govern autonomous workloads. You don't have to invent the ontology, curate the controls, or maintain the framework mappings — that foundation is published, usable, and continuously maintained.
Build on Apeiris if you are building…
Any product that needs a defensible, open control-and-evidence foundation instead of a proprietary one it has to invent and defend alone.
AI governance and GRC tools
Map your customers' controls to a public, versioned control set with framework provenance already attached — NIST AI RMF, ISO 42001, the EU AI Act, and more.
Runtime assurance and agent security
Turn detections into control semantics: each Apeiris control carries a blocking effect and the evidence it requires, so a runtime signal maps to a governed decision.
Audit and continuous compliance
Build on the Evidence Proof Map and the attestation ontology — obligations resolve to addressing controls, required evidence, and an honest coverage verdict.
Procurement and third-party review
Assess a vendor's AI system against the same open domains and evidence requirements every party can read, cite, and recompute — no black-box scoring.
Agent platforms and registries
Query the corpus from your agents over a single governed MCP server, with provenance on every response and a signed manifest behind it.
IDE plugins, MCP servers, and dev tools
Pull controls, obligations, and the graph straight into the developer's workflow. The API is the data; the schemas are versioned and stability-contracted, and public.
What you get.
An open substrate, published free under CC BY 4.0 — free to use, cite, redistribute, and build commercial products on.
- The full control corpus
Every domain matrix as normalized JSON: controls, validation objectives, required evidence, blocking effects, and framework mappings with fit and citation basis.
- A public conformance contract
The identifier grammar, every controlled vocabulary, the evidence ontology, and the consumer invariants — the ATT&CK-style data dictionary, derived live from the data.
- MCP servers for agents
One governed, read-only server exposes the corpus, graph, evidence, obligations, and a deterministic reasoning layer — checksum-verified on load.
- A knowledge graph
Typed nodes and provenance-carrying edges — control → threat → framework clause → normative source — deep-linkable and queryable.
- The Evidence Proof Map
Per-obligation proof chains for the anchored coverage frameworks: obligation → addressing controls → evidence → verdict and gap.
- A signed, verifiable manifest
Ed25519 over the JCS-canonical bytes, with a per-artifact SHA-256. Don't trust us — recompute every hash yourself, in the browser.
The pattern: fetch → verify → use.
Three steps, no login, nothing to install. Everything is static JSON behind a signed manifest.
Fetch
Pull the manifest.json and the artifacts you need — the domain matrices, the graph, the proof map — straight over HTTPS.
Verify
Check each artifact's SHA-256 against the manifest, and the manifest's Ed25519 signature against the published trust anchor. Trust the bytes, then use them.
Use
Read controls, obligations, evidence, and relationships — in your app, your pipeline, or your agent over MCP. mapped ≠ satisfied stays explicit.
Run it in fifteen minutes.
Not another architecture diagram — a single zero-dependency file that does the whole pattern: verify the signed manifest, verify an artifact's hash against it, resolve one framework obligation to its addressing controls, and render the evidence each one requires.
$ node prove-obligation.mjs
Node 18+ only (global fetch + node:crypto). Full source: /examples/prove-obligation.mjs. Set OBLIGATION=EU-AIA-ART09-01 to try another, or APEIRIS_BASE=… to run against a local copy. See a whole action on the wire — action-event → evidence-packet → decision (what they are).
✓ manifest signature verifies against ed25519-1b2d2270ebb2b50d — corpus release 2026.07.16 ✓ proofmap.json integrity — sha256 c225275b89eec290… matches the signed pin Obligation EU AI Act · Art. 5(1)(a) — Prohibited AI — subliminal manipulation Coverage direct · 4 addressing control(s) EF-05 · EU AI Act Prohibited Practices Governance (Art. 5) [fit:direct · basis:anchored] objective Every AI system … must have an Art. 5 screening record … evidence • prohibited_practices_register version-controlled with last_updated_date … evidence • art_5_screening_record per system: screening_date, register_version_used … … mapped ≠ satisfied — this is the evidence you would need to PRODUCE; it does not assert you hold it.
A foundation you can stand on.
Infrastructure has to be trustworthy and stable. Apeiris is built to be both — openly.
Signed and recomputable
A signed manifest and per-artifact hashes mean a consumer can prove the corpus wasn't tampered with — without trusting Apeiris. The trust story, in full.
StabilityVersioned releases
A CalVer corpus version and a machine-readable changelog let you pin to a release and detect drift. Schemas are public and evolve deliberately.
ProvenanceCited, not asserted
Every framework mapping carries its fit and a computed basis — anchored (validated against captured primary text) or asserted. Honest by construction.
Building something on Apeiris?
We are building this as shared infrastructure, and it is free during beta. If you're building a governance, runtime, procurement, audit, or agent-security product on the corpus, we'd like to hear about it — and help.