Skip to content
beam-agents
GitHub

7 pages

Learn

What the runtime is, how an activation works, and the invariants it holds.

  • A runtime that turns an agent into a keyed, stateful, fault-tolerant Beam transform — not a framework for writing agents.

  • InstallPartial

    How to get beam-agents today — from source, at a git ref, because no release has been published to PyPI.

  • The dataflow shape, the two execution paths through RunAgent, and why iterative loops cycle through the message bus instead of the DAG.

  • The seven rules the implementation is held to, what each one buys, and where each is tested.

  • What the runtime stores per key, how it is bounded, when it is collected — and the durable long-term tier behind it.

  • Write an agent, wire it into a pipeline, and run it on the DirectRunner with no credentials.

  • GlossaryStable

    The terms this project uses precisely — activation, entity key, staging, continuation, intent, effector, re-injection, replay cache, seq.