Claw Orchestrator

Claw Orchestrator is a self-hosted runtime for developers who want to coordinate coding agents on their own machine or server. It brings Claude Code, Codex, Antigravity, Grok Build and OpenCode into one interface, with support for custom coding CLIs. The runtime and browser dashboard run locally.

Sessions retain context across requests, and a conversation can move to another engine or model in the same workspace. For work that benefits from multiple agents, councils compare answers and vote toward agreement while agents work in separate Git worktrees. A simpler parallel mode collects independent answers with optional synthesis. Autonomous loops assign planning, coding and review to independently chosen engines and models.

Workflows save their progress so they can resume after a restart. Verification goes beyond an agent's own claim of success: the runtime can check tests, HTTP responses, screenshots and file changes against acceptance criteria. Runs with those criteria must pass before completion; runs without them are marked unverified. A durable activity ledger tracks token use and cost, and the runtime enforces spending caps.

It's open source under MIT. An OpenAI-compatible API connects existing chat clients to its coding sessions, while MCP exposes its tools to hosts such as Claude Desktop and Cursor. ACP lets editors including Zed and JetBrains use it as their coding agent. Its Ultraapp workflow uses an Opus council to build and locally deploy a web app from a structured interview.