Agentic Platform: Open-Sourcing the Agentic Identity Broker

Today we are open-sourcing the Agentic Identity Broker, an MIT-licensed component that lets AI agents act on behalf of users across third-party systems without ever holding the users' provider tokens. We built it while working on our own agentic platform, where we ran head-first into two questions: how to handle agentic identity, and how to give agents reasonable access across multiple systems. In Agentic Engineering at Zalando: a snapshot, Bartosz Ocytko described our platform work at a high level. In this post, we share details how we are building that platform from open-source components for the agent runtime and AI gateway, and why we built one core piece ourselves: an identity broker that captures delegation chains and enables access to third-party services.

Agentic Identity: Old Problems Magnified¶

When we look at agentic identity, we need to solve three core problems: (a) how to identify agents through their own machine identity, (b) how to give agents access to a multitude of third-party systems that each issue their own access tokens, and (c) how to make access decisions based on both the users' and the agents' permissions.