# AgentID

> AgentID is the sign-in button for AI agents. It gives every agent its own verified identity, backed by a real email address, so agents can sign in to apps without borrowing their owner's account. Apps that add AgentID always know which human is behind each agent.

AgentID is a product of AgentMail (YC S25). It is a standard OpenID Connect provider designed specifically for AI agents. Any app that already accepts Sign in with Google can accept AgentID with two configuration values and no code changes.

## The problem AgentID solves

AI agents today sign into apps using their owner's credentials. Apps receiving agent signups today have no way to tell whether hundreds of new accounts represent one person's agent fleet or many real customers. This creates three problems:

1. Apps cannot distinguish real customers from agent-created noise
2. Apps cannot enforce per-human rate limits, seat caps, or tiers
3. Apps with compliance obligations cannot verify the human behind an agent's actions

AgentID solves all three by passing a verified `owner_email` claim to registered apps on every sign-in.

## How AgentID works

When an agent signs in to an app that supports AgentID:

1. The agent clicks "Sign in with AgentID" on the app's login page
2. The agent authenticates against auth.agentid.com using OpenID Connect
3. The app receives an id_token containing the agent's identity: `sub` (stable subject), `email` (verified inbox address), and `email_verified`
4. If the app is a registered client and asked for the owner scopes, the id_token also carries the human owner's name and email (`owner_name`, `owner_email`), with `owner_email_verified: true` alongside `owner_email`. `/userinfo` returns the same values with the access token.

The signing key never leaves the browser that generated it. Each sign-in is a fresh signature over a one-time challenge. Owner scopes also need the `App: Share Owner` permission on the agent's AgentMail API key; without it, the agent can't finish the sign-in alone and the organization owner has to approve it.

## Client tiers

**Open clients** need no registration. The `client_id` is a URL the app controls. Open clients get the `openid`, `email` and `profile` scopes.

**Registered clients** are created in the AgentID console or with `npx @agentmail/agentid-cli init`. Registered clients get a client secret, up to 10 redirect URIs, and access to the `owner_profile` and `owner_email` scopes.

## Integration

Clerk ships AgentID as a built-in SSO connection (strategy `oauth_agentid`). AgentID also works as a custom OpenID Connect provider in Supabase, Auth0, and Auth.js v4. Better Auth apps instead install `@agentmail/agentid-better-auth`, an AgentMail-maintained provider helper for Better Auth's Generic OAuth plugin (Better Auth 1.7.2 or later). Run `npx @agentmail/agentid-cli init` from an existing app for browser-authorized registration and supported provider configuration. For apps that implement OIDC directly, the issuer is `https://auth.agentid.com`. Discovery, JWKS, PKCE with S256, and standard authorization code flow are all supported. Tokens are signed with ES256.

## Comparison to related products

- **Microsoft Entra Agent ID and Google Cloud Agent Identity**: enterprise IAM for agents inside one company's ecosystem. AgentID works across any app on the open web.
- **Auth0 for AI Agents (Okta)**: enterprise identity for agents inside customer applications. AgentID is a public identity provider, closer to Sign in with Google in scope.
- **Arcade**: focused on what agents can DO (permissions on actions), not who they ARE at sign-in. Complementary to AgentID.
- **WorkOS auth.md**: a protocol apps host describing how agents can register. AgentID can serve as an identity provider inside auth.md's verified flow.
- **Stripe Projects**: identity plus payments inside the Stripe ecosystem. AgentID is domain-agnostic.

## Key documents

- [How it works](https://www.agentid.com/#how-it-works): The sign-in flow
- [Security](https://www.agentid.com/security): Cryptographic details, trust boundaries, and revocation model
- [FAQ](https://www.agentid.com/#faq): Common questions from agent owners
- [Developer docs](https://www.agentid.com/docs): Full OIDC specification, endpoints, scopes, and token claims
- [CLI quickstart and troubleshooting](https://www.agentid.com/docs/cli): Detect, register, configure, verify, recover, unlink, and unregister AgentID from a project
- [Browser sessions](https://auth.agentid.com/sessions): Inspect and forget AgentID sessions saved in the current browser
- [Brand](https://www.agentid.com/brand): Logo assets and usage guidelines for Sign in with AgentID
- [Get started](https://console.agentmail.to/sign-up): Create an AgentMail inbox to get an AgentID

## About the makers

AgentID is a product of AgentMail (YC S25), the email infrastructure for AI agents. AgentMail provides programmatic inbox creation, send/receive functionality, webhooks, multi-tenant isolation, and scoped API keys. AgentMail is available at https://www.agentmail.to and its docs at https://docs.agentmail.to.

## Terminology

- **Agent**: an AI system that acts on behalf of a human, often with its own credentials and identity
- **Agent ID**: a general term for a unique identifier that lets an agent authenticate as a distinct entity
- **AgentID** (one word, capitalized): the specific OpenID Connect provider built by AgentMail
- **Owner**: the human who owns and is accountable for an agent's actions
- **Relying party (RP)**: an app that accepts AgentID sign-ins
- **owner_email**: the verified email address of the human behind an agent, passed to registered apps