# APEX PSI — Proof of Stateful Integrity

> Open-standard evidence protocol for AI governance. Records the existence and
> integrity of a decision, document or artefact at a point in time. It does not
> judge whether the contents are true, and verification is free forever.

Canonical site: https://ai-governance-standard.com
Full text for models: https://ai-governance-standard.com/llms-full.txt

## Verify without asking anyone

- npm: `npm install apex-psi-verifier` (zero dependencies; Node, Deno, Bun, browsers, CI)
- PyPI: `pip install psi-verifier`
- Both are published at **1.3.0** and both are MIT. A seal whose Ed25519 signature
  could not be checked is never reported as a pass.
- Reproduce our own conformance result: https://ai-governance-standard.com/conformance

A seal is an **audit trail**: it proves a statement was *made* at a timestamp and has
not changed since. It is not proof that the statement is true.

## Connect an AI agent (MCP)

Two doors over the same ledger. Verification through either is free and needs no payment; the
hosted door needs a free APEX sign-in, the npm door needs no account at all.

- **Install anywhere (stdio, npm, no account):** `npx -y apex-psi-mcp` — tools `seal`, `verify`,
  `anchor`, `cite`, `verify_chain`.
- **Hosted, no install (streamable-http, OAuth):** manifest at
  https://www.ai-governance-standard.com/.well-known/mcp/server.json — tools `seal`,
  `verify_hash`, `anchor_status`, `list_attestations`, `ledger_stats`, `protocol_info`. An
  anonymous call returns HTTP 401 with a `WWW-Authenticate` resource-metadata pointer.
- OAuth metadata: https://ai-governance-standard.com/.well-known/oauth-authorization-server
- Free allowance: 20 operations per minute, 100 per day, enforced per server instance.
  A higher tier buys capacity, never a finding.

## Core pages

- /protocol — protocol specification
- /spec — EU AI Act Article 50 mapping
- /standard — Compliance-Receipt HTTP header
- /verify — public hash verification, no account
- /conformance — the test suite and its result
- /cite — how to cite a receipt, and this protocol
- /corrections — our own mistakes, by dated entry; pages are never silently edited
- /pramaan — human witness sealing
- /api — REST API reference
- /integrations — integration surfaces
- /ledger — company recomputation ledger
- /partners — verified, integrated and powered-by programmes
- /agent — AI agent run sealed as a step-by-step hash chain (sign-in)
- /media — browser-side file fingerprint sealing with an informative C2PA-shaped hash record
- /console — signed-in list, search and JSONL export of your own receipts

## Standards and algorithms

RFC 8785 (JCS), SHA-256 (FIPS 180-4), Ed25519 (RFC 8032), ML-DSA-65 (FIPS 204),
LMS-W4-SHA256 (NIST SP 800-208). Canonical seal schema: PSI-SEAL/1.0.0.

Five PSI documents are **filed Internet-Drafts** (Individual Submissions, status
Informational, author K. Singh). **None is an approved standard and none has been
adopted by a working group.** Measured 21 October 2026, each revision below returns
HTTP 200 from the IETF archive:

- draft-singh-psi-02 — expires 16 April 2027
- draft-singh-psi-http-00 — expires 16 April 2027
- draft-singh-psi-agent-01 — expires 16 April 2027
- draft-singh-psi-merkle-01 — expires 16 April 2027
- draft-singh-psi-post-quantum-00 — expires 19 January 2027

Rule we hold ourselves to: a draft is called filed only when its datatracker URL
returns 200. https://datatracker.ietf.org/doc/draft-singh-psi/

## Trust anchor

https://ai-governance-standard.com/.well-known/apex-psi-trust-anchor.json

License: MIT