The open identity, provenance, and machine-resolution standard for AI agents.
Every production AI agent should have an identity: a permanent agenid:<ULID>, an accountable operator, declared purpose and disclosure posture, and verification records that anyone — a customer, an auditor, or another agent — can check without trusting AgenID's own database.
| 📜 Specification | AgenID v1.1.1 Final Protocol Specification — normative. MIT. |
| 🧩 JSON Schemas | manifest, manifest-proof, assertion, keys, authorities — draft 2020-12 |
| 🔏 Test vectors | Real RFC 8785 canonical bytes and Ed25519 signatures, reproducible from public seed labels |
🛠 @agenid/core |
TypeScript reference implementation — identifiers, canonicalization, schemas, proof engine |
Agent identity agenid:<ULID> — permanent, non-semantic, never re-keyed by platform or deployment
Manifest operator's DECLARED self-description; plain canonical JSON; never signed directly
ManifestProof operator key → DECLARED VerificationAssertion authority key → VERIFIED
Signing Ed25519 (pure) over RFC 8785 canonical bytes of the object minus "signature"
Keys agenid:key:<ULID>, role ∈ {operator, authority}, two-path discovery that must agree
Important
DECLARED never implies VERIFIED. VERIFIED never implies AUTHORIZED. Platform configuration implies none of them.
Specification v1.1.1 Final (Erratum E1 applied). Levels L1–L4 issuable; L5 reserved. Root authority key ceremony pending — no production assertion is trustworthy end-to-end until the pin is published in the spec repository.
Maintained by AI Venture Holdings LLC. Issues and proposals: open one on the spec repository.