Learn how agents work, one step at a time

Each explainer is a diagram you advance with Next step. It follows one piece of work through the system, so you can stop on the frame that puzzles you and look as long as you like.

  1. The full journey of one task in PugPort

    From the message you type to the answer you get back: how a task is read, routed, run, handed over when an account runs dry, and finished.

  2. How PugPort picks an account and a model

    Usage left on each account, how hard the task is, and the model and effort that fit it. Why the same request can land in different places.

    Coming soon
  3. Inside Pi: anatomy of an agent harness

    Based on the MIT-licensed Pi project by earendil-works: the loop, tools, context and session that turn a model into an agent.

    Coming soon