Skip to main content
Free & Open Source

Governed delegated access for AI agents

Your assistant reads tickets, reviews pull requests, and sends email for your people. OAuth lets it connect, but it cannot tell one agent from another, ask about one specific email, or take back one agent's access. The broker decides what each agent can do for each person, keeps the provider tokens out of the agents, and gives every user one place to approve and revoke access.

Get startedWhy does this exist?
Consent screen for delegating access to an AI agent
What this is

Identity infrastructure for agents that act on a user's behalf

These docs focus on the concerns an IAM team or platform operator cares about: the delegation model, security posture, deployment, configuration, and the API contracts other systems integrate against.

Delegated access, governed

A user grants an agent access to specific third-party services at a chosen permission set, with optional expiry. One grant per user and agent, revocable at any time.

An encrypted token vault

The broker stores third-party access and refresh tokens encrypted at rest with envelope encryption bound to each service, so agents never hold long-lived provider credentials.

Token exchange at the edge

A gateway swaps an agent token for the right third-party token per request using RFC 8693, enforced by policy — transparently, with an optional Envoy sidecar.

Start in the right place

Read by task, not by internals

Introduction

What the broker is, the delegation problem it solves, and when to reach for it.

Open sectionConcepts

The delegation model, architecture, server modes, token exchange, and encryption.

Open sectionGet started

Run the full stack locally and walk a delegation from consent to token exchange.

Open sectionAPI reference

The end-user and admin OpenAPI contracts, rendered from the source specifications.

Open section