gitmyhub

ChatDev

Python ★ 34k updated 24d ago

ChatDev 2.0: Dev All through LLM-powered Multi-Agent Collaboration

A platform to build and run teams of AI agents that work together on complex tasks through configuration, without writing code.

PythonLLM APIsWeb frontendReinforcement learningsetup: moderatecomplexity 4/5

ChatDev is a platform that lets you build and run teams of AI agents working together to accomplish complex tasks — without writing any code yourself. Think of it as assembling a virtual workforce where each worker is an AI model, and you define the roles, workflows, and tasks through simple configuration files rather than programming.

The project has two major versions. The original ChatDev 1.0 acted like a virtual software company where AI agents played roles such as CEO, programmer, and tester, collaborating to design, write, and document software automatically. Version 2.0, called DevAll, expands this concept into a general-purpose platform: you can orchestrate agents for tasks far beyond coding, such as data visualization, 3D content generation, and deep research.

Under the hood, agents communicate with each other through structured conversations. The platform coordinates which agent speaks when, what information gets passed between them, and how the overall workflow progresses toward a goal. Recent research baked into the project includes a reinforcement-learning-trained central orchestrator that dynamically decides which agents to activate and in what sequence, reducing unnecessary computation while improving output quality.

You would use ChatDev if you want to automate a multi-step process that benefits from having specialized "workers" each handling a portion of the problem — for example, one agent researching, another summarizing, a third formatting the final output. It is especially suited for builders or researchers who want to experiment with multi-agent AI systems without building the coordination infrastructure themselves.

The stack is Python on the backend, with a web-based frontend for the 2.0 platform. It integrates with large language models as the intelligence behind each agent and supports running the whole system locally or through a hosted service.

Where it fits