Browse Code
ZeroID is AI agent identity infrastructure: it keeps AI agent identity separate from human identity and access control. Human IAM assumes a person behind every credential. Autonomous agents don't qualify. They act with no human in the loop, spawn sub-agents, and run at machine speed.
ZeroID is agentic identity management built for non-human identities: cryptographically verifiable credentials, workload attestation, delegated authority across agent chains, and real-time revocation. Built on OAuth 2.1, WIMSE/SPIFFE, and RFC 8693, the emerging standards for AI agent identity security.
Each agent gets a stable, globally unique identity URI, its own principal rather than a shared service account or a borrowed human credential. Delegation attenuates scope automatically: a sub-agent inherits only what the orchestrator already holds. Every token carries the full on-behalf-of chain, so all programmatic access is governed, auditable, and attributable, cryptographically.
Features
- Per-agent cryptographic identity — every agent gets a stable WIMSE/SPIFFE identity URI: its own principal, not a shared service account or a borrowed human credential.
- Delegated authority with automatic scope attenuation — RFC 8693 token exchange; a sub-agent inherits only what the orchestrator already holds, capped by its own policy, with delegation-depth tracking and enforcement.
- Full on-behalf-of chain in every token — who authorized it, what scope was granted, how deep the delegation goes. Introspection returns the complete chain, so every action is attributable.
- Real-time revocation with cascade — revoking an upstream credential invalidates everything issued beneath it.
- Human-in-the-loop approval — CIBA backchannel approval with poll, ping, and push delivery for actions that need sign-off.
- Attestation framework and sender-constrained tokens — pluggable attestation verifiers, single-use nonce-bound WIMSE proof tokens, DPoP (RFC 9449).
- Credential policy engine — enforce TTL, permitted grant types, trust levels, and max delegation depth per agent.
- Continuous access evaluation — OpenID Shared Signals Framework and CAEP, so access is revoked on signal rather than at token expiry.
