What is Apeiris?

Apeiris gives enterprise AI systems a shared evidence model for verifying who authorized an AI action, what data and knowledge it used, what it actually did, and whether the effect stayed within policy.

Today, Apeiris is an open, signed corpus of controls, evidence requirements, relationships, and framework mappings — with interactive analysis and assessment tools. It is designed to become the evidence substrate for systems that evaluate consequential AI actions; the empirical runtime evaluation platform is in development. This page states which is which, plainly.

Four things, one name — kept distinct.

Apeiris blends easily into one idea. It is actually four related things at different stages. Reading them apart is the fastest way to understand what you can use today.

Apeiris Open Evidence Corpus

Available now

The published, signed data: 641 machine-readable controls across twelve domains, framework mappings, evidence requirements, the relationship & provenance graph, proof obligations, and a signed manifest — free under CC BY 4.0.

Includes: the twelve domain matrices · the knowledge graph · the Evidence Proof Map · normative source registry · MCP access · the in-browser integrity verifier.

Browse the domains →Fetch the artifacts →Verify the signature →

Apeiris Evidence Fabric

The architecture

The interoperability model through which evidence about an autonomous action can be expressed, related, verified, and composed — a shared ontology, stable namespaces, cross-domain relationships, and an attestation model, so proof obligations travel between the frameworks you already answer to.

This is the design, not a hosted service: the schemas, the identifier grammar, the evidence ontology, and the composition rules that let the corpus behave as one fabric.

Compositional assurance →One action, fully traced →

Apeiris Advisor

Available now · beta

A browser-based planning and self-assessment tool over the open corpus. You declare your stack; it returns a maturity read, framework coverage, and a phased roadmap — computed in your browser from your own answers.

It is a planning aid, not an empirical assessment, audit result, or compliance determination — it never inspects your environment.

Run the Advisor →

Apeiris Platform

In development

The hosted capability that will obtain real action context, evaluate an action's proof obligations against evidence from your systems, maintain assurance state, and produce reviewable, empirical attestations.

Not equivalent to the published corpus. Connectors, runtime evidence collection, and action-level evaluation are being built; nothing on this site claims Apeiris observes or evaluates your real actions today.

What is verifiable today →

Available today vs. being built.

The single boundary the whole site turns on.

Available today

The full corpus — 641 controls, evidence requirements, blocking posture
Framework mappings with fit, basis, and relationship typing
The knowledge graph and the Evidence Proof Map
Signed, versioned releases you can verify byte-for-byte
Browser tools: Advisor, Prove, Verify, Graph, Analysis
MCP access and the consumer SDK over the same signed artifacts

Being built

Connectors that collect real runtime evidence from your systems
Action-level evaluation of proof obligations against that evidence
Assurance state and empirical attestation production
The hosted platform and its enterprise integrations

What Apeiris is not.

Stated as plainly as what it is.

Not a runtime that observes or evaluates your real AI actions today — that is the platform, in development.
Not a claim to prove the model behaved correctly — Apeiris evaluates evidence about the action, not the model.
Not a replacement for your controls — it sits over authentication, retrieval, policy, and approval, evaluating the interaction between them.
Not a compliance certification — a signature proves the bytes are authentic and unchanged, not that a mapping or interpretation is correct.
Not a decision gate that unilaterally declares an AI "safe" — it produces auditable evidence, advisory and replaceable.

Where to go next.

The idea