ChatDev 2.0 (DevAll) is a multi-agent orchestration platform in which you define agents, workflows, and tasks through configuration rather than code. It grew out of ChatDev 1.0, a virtual software company where role-playing agents (CEO, CTO, programmer, and so on) collaborate through a chat chain to design, code, test, and document software. That version lives on a legacy branch, and 2.0 generalizes the approach to scenarios such as data visualization, 3D generation, and deep research.
The platform is a Python backend (FastAPI) paired with a Vue 3 web console. Workflows are YAML files that reference model credentials through variable placeholders. A validation target checks them for syntax and schema errors, and a sync target uploads them to the database. The console has a tutorial, a drag-and-drop workflow canvas for configuring nodes and context flows, and a launch view for running workflows, watching logs, inspecting intermediate artifacts, and giving human-in-the-loop feedback.
For automation, a Python SDK (runtime.sdk.run_workflow, also published as the chatdev package) executes a YAML workflow with a task prompt, attachments, and variable overrides and returns the final node message. OpenClaw can drive ChatDev through an installable skill, either invoking existing agent teams or creating new ones from a chat request. Developers extend the platform with new nodes, providers, and tools across the server, runtime, workflow, and entity modules.
Research lines from the same group feed into the codebase on separate branches: MacNet (directed-acyclic-graph collaboration among more than a thousand agents), a puppeteer-style orchestrator trained with reinforcement learning (NeurIPS 2025), and Experiential Co-Learning and Iterative Experience Refinement for reusing shortcut experiences across tasks.
Features
- Configuration-driven agents: agents, workflows, and tasks are declared in YAML with no application code
- Visual workflow canvas: node parameters and context flows are edited by drag and drop in the web console
- Run monitoring: real-time logs, intermediate artifacts, and human feedback during execution
- Python SDK: run_workflow runs a YAML workflow programmatically with prompts, attachments, and variables
- OpenClaw integration: a skill lets OpenClaw invoke or create ChatDev agent teams
- Provider flexibility: an API key and base URL select any compatible LLM provider
- Docker Compose: backend and frontend run as containers with live reload of local changes
- Extension points: new nodes, providers, and tools slot into the modular runtime
- Legacy software-company mode: ChatDev 1.0 with chat-chain phases, Git mode, Art mode, incremental development, and human-agent interaction remains on its own branch




