Claude Code's source just leaked — I extracted its multi-agent orchestration system into an open-source framework that works with any LLM

A significant leak of Anthropic's Claude Code project, exposing over 500,000 lines of TypeScript source code, has provided an unprecedented look at its sophisticated internal architecture. Developer JackChen02 focused on the core multi-agent orchestration system, which includes components like a goal-decomposing coordinator, a team management system, and a task scheduler. He has meticulously re-implemented these design patterns from the ground up to create 'open-multi-agent,' a clean, standalone framework. The project is fully open-source under an MIT license and written in TypeScript, offering a production-ready blueprint for complex AI agent workflows without any copied proprietary code.

Unlike the original Claude Agent SDK, which spawns separate CLI processes for each agent, open-multi-agent runs entirely in-process, making it highly portable for serverless functions, Docker containers, or CI/CD pipelines. The framework is fundamentally model-agnostic, enabling developers to build teams of agents that can mix and match different LLMs, such as Claude 3.5 Sonnet and OpenAI's GPT-4o, within a single coordinated system. Key implemented patterns include a TaskQueue with topological dependency resolution to manage complex task sequences and a defineTool() function with Zod schema validation for robust tool creation. This provides a significant leap in accessible tooling for developers looking to build advanced, multi-step AI applications.