OPEN PROTOCOL · DEVELOPER TOOLS

The Open Trust Layer
for AI Agents.

ATP is an open effort to make agent identity, trust evidence, and privacy controls portable across systems. Explore the specifications in development, build with available tools, and help test interoperability.

Start building →Read the specifications
did:atp identityVerifiable trust evidenceMCP · A2A · moreOpen SDK
●   ●   ●agent.mjs · SDK 6.0.2
import { Agent } from 'atp-sdk';

const agent = await Agent.quickstart('MyBot');
console.log(agent.getDID());
console.log('Standalone:', agent.isStandalone());
Runs standalonenpm install atp-sdk@6.0.2

THE CORE IDEA · 01 / 04

Agents move.
Trust should travel with them.

An agent can cross tools, companies, and protocols in a single workflow. ATP is developing portable ways to identify an agent and assess verifiable trust evidence at each boundary.

The available SDK gives developers a starting point today. The broader privacy and interoperability work is still in development.

THE OPEN TRUST LAYER · 02 / 04

Four connected building blocks.

Explore the specifications in development and the available implementations behind ATP.

01 / ATP

Identity

Give an agent a portable did:atp identity backed by hybrid Ed25519 and ML-DSA-65 key material.

Specification: Unofficial draft input
Implementation: Reference code and vectors
Explore draft →02 / ATP

Trust

Evaluate verifiable evidence and bounded trust levels. Issuer authority and credential status remain under review.

Specification: Unofficial draft input
Implementation: Partial implementation
Explore draft →03 / ATP

Privacy

Design interactions around pairwise identifiers and selective disclosure. Zero-knowledge work remains experimental.

Specification: Unofficial draft input
Implementation: Partial implementation
Explore draft →04 / ATP

Interop

Bring identity and trust to existing agent systems, including MCP and A2A, through open specifications and conformance work.

Specification: Unofficial draft input
Implementation: Cross-protocol work ongoing
Explore draft →

ATP draft input documents are shared for review with the W3C Agent Trust Protocol Community Group. These are unofficial draft inputs, not adopted Community Group Reports or W3C Standards. W3C does not endorse this work. Implementation support varies by capability.

FOR DEVELOPERS · 03 / 04

Four ways to build.

Use the SDK in an existing application, scaffold a project, guard Claude Code or Hermes from the terminal, or inspect the available conformance suites.

SDK

Add identity and tool permission decisions to an existing JavaScript or TypeScript agent.

npm install atp-sdk@6.0.2Read the SDK quickstart →

Starter

Create a runnable JavaScript starter using SDK 6. Its optional screen saves settings; it does not enforce policy.

npx create-atp-agent@1.7.0 my-agent --no-dashboardRead the starter guide →

Terminal

The atp command. Checks Claude Code and Hermes tool calls against an ATP profile, manages a local did:atp identity and records decisions, signed when an identity is configured. It can only tighten Claude Code's own permissions.

npm install -g @atpdeveloper/atp-cli@0.1.1Read the atp-cli guide →

Conformance

Run the available suites from a repository checkout with dependencies installed. This does not certify every draft capability.

npm run conformanceInspect the test runner →

Use Node.js 20 or newer; the local quickstarts need no ATP account or model API key. For Python, DID primitives, Profiles, and Hermes, compare all Core packages →

VENTO CORE 1.3 · TRUST FLOW

Know what an agent can do.

An agent's identity, evidence, and granted authority should remain visible at the moment of action. This is a design illustration of the trust boundary; it is not a live decision or a claim of conformance.

Explore developer tools →

Want to inspect a real policy boundary? Run the action-boundary example →

ILLUSTRATIVE AGENT ACTION

A request with a clear boundary

Read + propose
  1. 01
    Identify the agent

    Resolve its portable identity and verify the presented evidence.

  2. 02
    Check authority

    Evaluate scope, policy, and expiry for this specific request.

  3. 03
    Record the outcome

    Show the decision and evidence that a relying system can inspect.

Concept preview · capabilities depend on the implementation and policy.

ATP STUDIO · 04 / 04

Open foundation.
Enterprise control.

Studio is the separate enterprise product for governing agent identity, authority, policy decisions, and audit evidence in real operations. It builds on the ATP trust layer with managed workflows and controls for teams.

Explore Studio ↗

ILLUSTRATIVE TRUST DECISION · ATP STUDIO

IdentityVerified
Issuer authorityChecked
PolicyEvaluated
OutcomeAuditable

Illustrative workflow · No live customer data

OPEN WORK · OPEN PARTICIPATION

Help make trust portable
across the agentic web.

Start building →View the codeJoin the Community Group