AgIS (Agent Identity System) is EPICORTEK's open protocol and reference implementation for DNS-backed identity verification of AI agents and autonomous workflows. It turns agent identity evidence into deterministic trust decisions: allow, deny, or review.

Is AgIS an IETF standard?

No. AgIS is an Individual Internet-Draft submitted to the IETF on 29 June 2026. It has not been approved, adopted by a working group, or standardized. It is a protocol proposal in early stage.

Is AgIS a DID system or blockchain-based?

No. AgIS is not a DID system, is not blockchain-based, and does not use a wallet or distributed ledger. It uses DNS TXT records and cryptographic key binding (Ed25519) for verifiable agent identity.

What are AgIS trust levels?

AgIS reports trust levels 0–5. Level 0 means no usable identity evidence. Levels 1–3 add syntactic validity, DNS binding, and cryptographic verification. Level 4 adds active status confirmation. Level 5 adds freshness, replay protection, and local policy enforcement. The offline reference implementation test vectors reach Level 4.

What is an Agent Card?

An Agent Card is a structured JSON document that describes an AI agent's identity, public key, capabilities, delegation scope, and lifecycle status. AgIS evaluates Agent Cards as the primary identity evidence source.

How does AgIS use DNS?

AgIS uses DNS TXT records to bind an Agent Identifier to an Agent Card URL and JWK thumbprint. This DNS binding provides a verifiable, decentralized link between an agent's claimed identity and its cryptographic key, without requiring a central registry.

Internet-Draft · draft-ayoub-agis-agent-identity-system-00 · submitted to IETF 29 June 2026

AgIS — Agent Identity System

Open protocol and reference implementation for agent verification and governance.

AgIS turns agent identity evidence into enforceable trust decisions.

GitHub npm SDK @epicortek/agis-sdk-ts npm CLI @epicortek/agis-cli

23: Deterministic test vectors
0.35ms: Avg per offline verification
Ed25519: Signature algorithm
4: RFC profiles

◉ Note

Status: AgIS has been submitted to the IETF as an individual Internet-Draft (draft-ayoub-agis-agent-identity-system-00), submitted 29 June 2026. It has not been adopted by any IETF working group and is not an approved IETF standard. The TypeScript SDK and CLI are alpha-quality and are intended for protocol development and testing, not unreviewed production deployment. agent:// is not yet an IANA-registered URI scheme.

Developer entry points

All public artifacts. No login required to access any of these.

GitHub repository

Public

epicortek/agis

Source, reference implementation, CLI, test vectors, CHANGELOG

npm SDK

Alpha

@epicortek/agis-sdk-ts

TypeScript SDK — v0.3.0-alpha.3 — offline agent verification

npm CLI

Alpha

@epicortek/agis-cli

Command-line agent identity evaluation and verification

IETF Internet-Draft

Submitted

draft-ayoub-agis-agent-identity-system-00

Canonical specification — submitted 29 June 2026

Test vectors

23 passing

23 deterministic — 0 failures

All components and selected negative cases covered

Formatted draft

On this site

epicortek.com/en/agis/draft

Same specification content, styled for readability

30-second install

The SDK is published to npm. No build step, no API key, no external service.

terminal

npm install @epicortek/agis-sdk-ts

TypeScript

import { signAgentCard, verifyAgentOffline } from "@epicortek/agis-sdk-ts";

// 1. Sign your agent card
const signedCard = await signAgentCard({ agentCard, privateJwk, keyId });

// 2. Verify offline — no API call needed
const result = await verifyAgentOffline({
  dnsTxtRecord: "agis=0.2.2; agent=agent://example.com/worker; card=https://...; jkt=...; card_sha256=...",
  signedAgentCard: signedCard,
  statusDocument: { agent_id, status: 
    "active", updated_at: new Date().toISOString()}
});

// result.decision:   "allow" | "deny" | "review"
// result.trustLevel:  0–5
// result.validIdentity: true | false

Verification flow

Each step adds a layer of evidence. The trust level at the final step determines the allow / deny / review decision.

Agent request

Agent Identifier + Agent Card URL presented
DNS TXT lookup

DNS record binds agent identifier to card URL and JWK thumbprint
Agent Card fetch & canonicalization

RFC 8785 canonical JSON — hash verified against DNS TXT card_sha256
Signature verification

Ed25519 (RFC 8037) + HTTP Message Signatures (RFC 9421) + Content-Digest (RFC 9530)
Lifecycle status check

Status document: active / suspended / revoked / compromised
Trust decision

allow · deny · review — deterministic, reproducible, auditable

Trust levels

AgIS reports trust levels 0–5. Higher levels reflect more verification evidence. The current offline identity verification test vectors reach Level 4; high-assurance signed request verification can reach Level 5 when freshness, replay protection, and local policy acceptance are enforced.

No identity evidence

deny

No usable identity evidence present. Request cannot be authenticated.

Syntactically valid

review

Agent Identifier and Agent Card are syntactically valid. Structure verified, not cryptographically bound.

DNS binding verified

review

DNS TXT record links the Agent Identifier to the Agent Card URL. Domain control confirmed.

Cryptographically verified

allow

Card hash (RFC 9530 + RFC 8785), JWK thumbprint (RFC 7638), and Ed25519 signature verified.

Active status confirmed

allow

Status document checked and agent is active (not revoked, not suspended, not compromised). Offline reference implementation test vectors reach this level.

Freshness, replay protection, and policy enforced

allow

Freshness window (max_age) enforced, replay protection (nonce / two-phase commit) verified, and local policy explicitly accepted. Required for high-assurance signed request verification.

What AgIS is

AgIS is EPICORTEK's open protocol specification and reference implementation for agent verification and governance. It defines how AI agents, autonomous systems, and non-deterministic workflows prove who they are, validate the authenticity of their requests, and operate under enforceable policy decisions.

The core mechanism: AgIS evaluates agent identity evidence from multiple verification sources and produces a deterministic operational decision — allow, deny, or review. These decisions are reproducible, auditable, and policy-bound.

AgIS uses infrastructure that already exists — DNS, HTTPS, and cryptographic primitives. No new PKI. No blockchain. No central registry.

The problem AgIS addresses

AI agents and autonomous systems are increasingly capable of taking real actions — making API calls, executing tasks, delegating to sub-agents, accessing data, and triggering workflows. But there is no standardized way for these systems to:

Prove who they are in a verifiable, machine-readable way
Demonstrate that a request came from the agent it claims to come from
Show the scope of their delegation authority
Signal that they are active (not revoked, not suspended)
Enable auditors and operators to verify what happened after the fact

Without this layer, trust in agentic systems is informal, unverifiable, and inconsistent. AgIS proposes a structured, DNS-backed approach to this problem.

How AgIS works

A DNS TXT record binds an agent identifier to an Agent Card and signing key. A verifier inspects the DNS binding, canonicalizes the Agent Card (RFC 8785), verifies the Ed25519 signature against the DNS-bound key (RFC 7638 JWK thumbprint), validates the HTTP Message Signature (RFC 9421), checks the Content-Digest (RFC 9530), verifies request freshness, and checks lifecycle status.

Agent Cards

Structured identity claims attached to an agent, evaluated against registered evidence.

DNS TXT bindings

Domain-verified association between an agent identifier and a DNS record.

Key thumbprints

Cryptographic key fingerprints (RFC 7638) used to bind requests to agent identity.

Signed HTTP requests

HTTP Message Signatures (RFC 9421) validated against the agent's registered keys.

Request freshness

Two-phase nonce commit prevents replay of stale or captured requests.

Lifecycle status

Active, suspended, revoked, or compromised — checked after cryptographic verification.

Revocation

Status document endpoint checked; revoked agents receive a deny decision.

Scoped delegation

Single delegation tokens with explicit scope constraints.

Delegation chains

Multi-hop delegation with scope narrowing enforced at every link.

After evaluating the available evidence, AgIS produces one of three decisions:

Allow — sufficient evidence; proceed with the operation
Deny — insufficient or contradictory evidence; block the operation
Review — evidence incomplete; escalate for human or policy review

These decisions are deterministic: given the same evidence and policy configuration, the same decision is always produced. This is validated by 23 deterministic test vectors.

IETF Internet-Draft

draft-ayoub-agis-agent-identity-system-00

AgIS has been submitted to the IETF as an individual Internet-Draft. It defines a DNS-backed identity and verification profile for AI agents using existing web infrastructure. No new PKI. No blockchain. No central registry.

Submitted 29 June 2026. This is an individual submission for community review — it has not been adopted by any IETF working group and is not an approved IETF standard.

View on IETF Datatracker ↗ Raw .txt ↗ Read formatted on this site →

What exists today

Internet-Draft

Submitted

draft-ayoub-agis-agent-identity-system-00 — submitted 29 June 2026. Individual submission to IETF.

View

TypeScript SDK

Alpha · npm

@epicortek/agis-sdk-ts — v0.3.0-alpha.3 published to npm.

View

CLI

Alpha · npm

@epicortek/agis-cli — alpha companion for evaluation and verification.

View

Deterministic test vectors

23 passing

23 test vectors covering all components and selected negative cases. All passing.

View

What AgIS is not

Understanding the precise scope of AgIS prevents misapplication.

Not a DID system

AgIS does not use or require Decentralized Identifiers. It operates on its own DNS-backed evidence model.

Not decentralized identity

AgIS does not implement DIDs, verifiable credentials, or W3C DID methods.

Not a wallet system

AgIS does not store credentials, keys, or identity material. It evaluates evidence.

Not blockchain-based

AgIS has no dependency on distributed ledgers, blockchains, or consensus protocols.

Not an IANA-registered scheme

agent:// is not yet a registered IANA URI scheme.

Not an approved IETF standard

The Internet-Draft has not been adopted by any IETF working group.

Not production enterprise software

The SDK and CLI are alpha. Not for unreviewed production deployment.

Not a replacement for ANS

AgIS addresses a different problem domain from Autonomous Naming Systems.

Not an agent framework

EPICORTEK does not replace orchestration, workflow, or model tooling. AgIS focuses on identity, policy, evidence, observability, and audit around agentic systems.

Status and maturity

Artifact	Status	Notes
Internet-Draft specification	Submitted (29 June 2026)	draft-ayoub-agis-agent-identity-system-00; not adopted by IETF WG
TypeScript SDK (@epicortek/agis-sdk-ts)	v0.3.0-alpha.3	Published to npm
CLI (@epicortek/agis-cli)	Alpha	Published to npm
Deterministic test vectors	23 passing	All components + selected negative cases
Agent Card evaluation	Implemented (alpha)	Core verification mechanism
DNS TXT bindings	Implemented (alpha)	Domain-based identity binding
Ed25519 signatures	Implemented (alpha)	RFC 8037; no symmetric secrets
HTTP Message Signatures	Implemented (alpha)	RFC 9421 profile
Content-Digest	Implemented (alpha)	RFC 9530
Delegation chain validation	Implemented (alpha)	Multi-hop with scope narrowing
Live DNS/HTTPS resolver	Planned (beta)	Reference implementation is offline-only
IETF standardization	Submitted (29 June 2026)	Individual Internet-Draft; not a WG item
Production deployment	Not claimed	Alpha reference implementation only