Repolode
Submit
Enderfga

Enderfga/claw-orchestrator

A TypeScript runtime that runs Claude Code, Codex, Grok Build, OpenCode and other coding agent CLIs behind one API, with checks the runtime runs itself.

View on GitHubWebsite
Stars
601
Forks
104
Open issues
1
Last commit
3 days ago

What is it?

Claw Orchestrator is a headless runtime for coding agent CLIs. It keeps Claude Code, Codex, Antigravity, Grok Build, OpenCode and any Agent Client Protocol agent running as persistent sessions behind one interface, and it can be driven from a CLI, an MCP server, an ACP editor agent, an OpenAI-compatible endpoint, a TypeScript SDK or an OpenClaw plugin.

The main idea is verification. You declare acceptance checks such as commands, HTTP probes, screenshots or file assertions, and the runtime runs them and reads the exit codes itself. A run ends as verified, refuted or unverified, so an agent cannot mark its own work as passing. Workflows are checkpointed, so runs can resume after a crash, and every turn is recorded in a ledger with model, tokens and cost, with an optional spend cap per run.

The project also supports fan-out of one task across several engines, moving a live session to another engine in the same workspace, and long-running Codex or Claude goals with resume after usage limits.

Key features

  • 01Runs Claude Code, Codex, Antigravity, Grok Build, OpenCode and ACP agents as persistent sessions.
  • 02Acceptance checks run by the runtime, producing verified, refuted or unverified verdicts.
  • 03Checkpointed workflows that resume, retry and cancel after a crash.
  • 04Per-turn cost ledger across engines and a per-run spend cap.
  • 05Fan-out to several engines, plus live session handoff between engines.
  • 06Available as a CLI, MCP server, ACP agent, OpenAI-compatible endpoint and TypeScript SDK.

Use cases

  • Checking that an agent's bug fix really passes tests the agent did not write.
  • Comparing answers from several coding agents on the same task.
  • Running a long Codex or Claude goal unattended with resume after usage limits.
  • Capping spend and tracking per-turn cost when many agent sessions run.

Quick start

  1. 1

    Install the CLI globally with npm

    npm install -g @enderfga/claw-orchestrator
  2. 2

    Start the local runtime and dashboard

    clawo serve &
  3. 3

    Run a task with checks on an engine

    clawo solve "The test fails. Fix price.js." --engine codex --check "npm test" --check "node ../holdout/price.spec.mjs" --wait
  4. 4

    Verify the run and list its turns

    clawo verify <runId>

Things to know

  • Requires Node 22+ and at least one engine CLI (claude, codex, agy, grok or opencode) installed and logged in.
  • Some engines need their own paid account or subscription, and free tiers can hit usage limits.

In collections

Multi-agent orchestration frameworksClaude Code tools, skills & pluginsOpen-source AI agent frameworks & appsAI coding assistants & agentsMCP servers

Topics

#ai-agent#agent-orchestration#coding-agent#mcp#multi-agent#claude-code#typescript#acp#agent-client-protocol#agentic-workflow

FAQ

What is claw-orchestrator?
A TypeScript runtime that runs Claude Code, Codex, Grok Build, OpenCode and other coding agent CLIs behind one API, with checks the runtime runs itself. Claw Orchestrator is a headless runtime for coding agent CLIs. It keeps Claude Code, Codex, Antigravity, Grok Build, OpenCode and any Agent Client Protocol agent running as persistent sessions behind one interface, and it can be driven from a CLI, an MCP server, an ACP editor agent, an OpenAI-compatible endpoint, a TypeScript SDK or an OpenClaw plugin.
Is claw-orchestrator free to use? What license does it have?
claw-orchestrator is open source under the MIT license. Check the license terms before using it in commercial products.
Is claw-orchestrator still maintained?
Actively maintained. The last commit to claw-orchestrator was on Nov 4, 2026.
What can I use claw-orchestrator for?
Checking that an agent's bug fix really passes tests the agent did not write. Comparing answers from several coding agents on the same task. Running a long Codex or Claude goal unattended with resume after usage limits. Capping spend and tracking per-turn cost when many agent sessions run.
How do I install claw-orchestrator?
Install the CLI globally with npm: npm install -g @enderfga/claw-orchestrator Start the local runtime and dashboard: clawo serve & Run a task with checks on an engine: clawo solve "The test fails. Fix price.js." --engine codex --check "npm test" --check "node ../holdout/price.spec.mjs" --wait Verify the run and list its turns: clawo verify <runId>
Where can I get claw-orchestrator?
The source code and installation instructions are on GitHub: https://github.com/Enderfga/claw-orchestrator

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects