Why not a traditional IdP?
If you use Keycloak, Auth0, Okta, or a corporate identity provider, you can expect an identity broker to compete with it. The broker does not compete with your IdP. It complements your IdP and depends on it. Your IdP answers one question: who is this human? The broker answers another: what can this agent do for this human with a third-party service?
Your IdP handles human login. The broker governs agent delegation. Both services receive requests in the same path but handle different concerns.
The IdP and the broker
An IdP authenticates people and federates user sessions. It handles login, single sign-on, multi-factor challenges, and the user directory. The broker does not provide these functions. It never authenticates a human. It trusts an identity from a proxy and decides which agents can act for the person. It also defines which third-party permissions an agent can use.
This division is intentional. The broker starts after authentication. It manages consent, holds third-party tokens in encrypted storage, and exchanges an agent token for a third-party token. It has no login page, password store, or user directory.
Division of responsibilities
| Your identity provider owns | The broker owns |
|---|---|
| Human login and credential verification | Per-agent, per-service delegation |
| Single sign-on across your applications | Consent expressed as permission sets |
| Multi-factor and step-up authentication | An encrypted vault of third-party tokens |
| The user directory and profile source of truth | RFC 8693 token exchange at the gateway |
| Password reset, social login, account lifecycle | Per-agent revocation and grant expiry |
The columns do not overlap. The IdP establishes identity. The broker governs the services and permissions that an agent can use.
They work together
The broker is behind an IdP-backed reverse proxy. The proxy authenticates each user,
often through your IdP. It sends the authenticated principal to the broker in
X-Remote-User by default. The broker accepts that header only from the proxy. The proxy
also enforces administrator privilege before requests reach the admin API.
The broker can validate a signed JWT from the proxy. It can read a display profile from the token claims. Your IdP or proxy issues that token. The broker does not run its own login. Human authentication stays in your existing stack.
When the broker fits
Choose the broker when
- An AI agent must act for a user with a third-party OAuth2 service.
- A user must be able to review, time-limit, and revoke per-agent, per-service consent.
- Third-party tokens must stay in one encrypted vault, not in agents.
- A gateway must exchange an agent token for the appropriate third-party token at request time. See token exchange.
You still need your IdP for
- Authenticating humans: login, passwordless, social, and enterprise credentials.
- Single sign-on across your applications.
- Multi-factor and step-up authentication.
- Being the source of truth for who your users are — their directory and profile.
The broker expects an existing authentication system. It does not replace an IdP, a high-throughput machine-to-machine authentication layer, a federation fabric, or an offline-verification system.
Related
- What is the Agentic Identity Broker? — the one-paragraph picture.
- Use cases — where delegated agent access is the hard part.
- Delegation and consent — the model the broker governs.
- Architecture — how the pieces fit at an operator level.