Open Agent Loops

The fundamental pieces to build and extend your own agent — your Jarvis, your Cortana, your Samantha.

A minimal, provider-agnostic agentic loop. Every fundamental piece sits behind an interface, so it can be swapped without touching the loop — the point of the package is to make the components of the agent plug-and-play, independently testable, and reliable.

Why It Exists

• Plug-and-play seams — the model client, memory, tools, and stop conditions are all interfaces. Implement them with a plain object or function.
• Streaming by default — ModelClient.stream() returns an async iterable of StreamEvents, so reasoning, text, and tool calls arrive incrementally.
• Bring your own front end — the loop is headless. It emits a typed AgentEvent stream via onEvent; render it to a console, the DOM, a TUI, or anywhere — the same run drives a CLI or a browser UI unchanged.
• Provider-agnostic — nothing in the core references a specific LLM SDK. Any OpenAI-compatible endpoint works with a raw fetch.
• Runs anywhere — the core has no platform APIs and a single dependency (zod), so the same ESM build runs in Node, Bun, Deno, and the browser.
• Independently testable — each seam is verified in isolation with deterministic test doubles; zero network.