m4trixDocs

Getting Started

What's Happening (Mental Model)

Here's the mental model for how a request flows through m4trix:

Flow Overview

  1. HTTP Request — A client sends a POST (or GET) to your exposed endpoint.
  2. expose() — Extracts the JSON payload, runs auth (if configured), and publishes a start event to the main channel.
  3. Main Channel — Events enter here. Agents subscribed to this channel receive the event.
  4. Agent(s) — Logic runs. The agent reads triggerEvent.payload, does work (e.g. calls an LLM), and emits new events.
  5. Client Channel — Emitted events are published to channels the agent is bound to. The client channel has an sse() proxy, so events are streamed as SSE to the browser.
  6. Browser — Receives SSE events and can render them in real time.

Key Ideas

  • Events are typed messages. Schemas are validated at runtime via Effect.
  • Channels route events. Agents subscribe to channels (input) and publish to channels (output).
  • Proxies determine how events leave the system — e.g. proxy.sse() for SSE, or a custom proxy kind of your own.
  • AgentNetwork wires everything: channels, agents, and the event plane.

Diagram

One request through the hello-world network: the browser POSTs to expose(), which publishes user-request on main; myAgent runs and emits agent-response events on client; they stream through proxy.sse() to the browser, followed by m4trix:run.completed.

Next