Write a plan. Watch your agents come alive.
Declare your agent topology in a `.wh` file. Apply it. Wheelhouse handles wiring, process management, and restarts.
Wheelhouse is the operating infrastructure for autonomous agent factories.
It lets you specify, deploy, monitor, and let your agents operate their own infrastructure — without human intervention.
The core idea
Section titled “The core idea”The agent is the operator.
Unlike existing orchestrators (LangGraph, CrewAI, AutoGen) — frameworks written by humans for humans — Wheelhouse is infrastructure operated by agents themselves. The .wh file is not static config — it is a living topology that agents read, modify, and apply as their needs evolve.
Key primitives
Section titled “Key primitives”| Primitive | Role |
|---|---|
| Stream | Real-time typed object bus — the connective tissue |
| Agent | Autonomous subscriber/publisher with a observe → decide → act cycle |
| Surface | Bridge between human users and a stream (Telegram, CLI, custom) |
| Skill | Versioned recipe stored in git, invoked via stream object |
| Cron | First-class scheduler that publishes CronEvent into streams |
.wh file | Declarative topology declaration — the Dockerfile of agentic infrastructure |
Why Wheelhouse?
Section titled “Why Wheelhouse?”Three things that don’t exist anywhere else:
- LLM-native IaC —
.whfiles are designed to be read and written by language models - Versioned communication contracts — every stream message has a typed, versioned Protobuf schema
- Autonomous observation→decision→action loop — the stream is the observation, compaction is the analysis, the
.whis the intent, apply is the action — all signed and audited in git