ChatDev 2.0: build multi-agent systems with zero code and develop anything, fast
Remember ChatDev (Chat-powered Software Development), the multi-agent framework that Andrew Ng strongly recommended, topped GitHub Trending for seven days straight, gathered 28k stars and 3,300 forks, and was the first to bring large-model agent collaboration to autonomous task solving?
Today, the joint team from Tsinghua University, Shanghai Jiao Tong University, OpenBMB, and ModelBest officially open-sources ChatDev 2.0 — a zero-code multi-agent tool anyone can use. Compared with version 1.0, which had multiple agents collaborate on a single software development task, ChatDev 2.0 can write code, make creative videos, produce academic reports, and more. Behind this greater capability: ChatDev 2.0 lets anyone create a wide variety of multi-agent systems for complex tasks through simple “drag + configure.”
This article includes video content. Please view it in the original WeChat post.
➤ Repository
🔗 https://github.com/OpenBMB/ChatDev
Agent development moves from complex code to visual interaction
In exploring and shipping multi-agent systems, developers often face a hard choice: put up with the rigidity and memory loss of low-code workflow platforms just for the convenience of drag-and-drop, or wrestle with complex, opaque code frameworks just to get real flexibility?
ChatDev 2.0 removes that choice — you can have both. It is not just another single agent framework; it is a fully open-source, zero-code productivity tool. Through an intuitive “Drag + Configure” paradigm, it turns opaque code logic into a visual interaction experience, bringing the barrier to building multi-agent systems close to zero and giving every idea its own “intelligent team.”
Build multi-agent systems with zero code — just drag and configure
Drag: drag freely on an infinite canvas to define collaboration edges between agents. Whether a simple linear flow or a complex mesh topology, a mouse-drawn connection is all it takes.
Configure: no more complex config files. Through form-based configuration, you can fine-tune each node’s role, prompt, base model, and context window.
From a simple article-writing assistant to complex collaboration flows that support 3D modeling, ChatDev 2.0 keeps leveling up with developers’ needs.
Underlying orchestration with exceptional freedom
ChatDev 2.0 is not just simple on the surface. Underneath, it implements multi-agent orchestration through a highly robust graph structure, giving developers exceptional freedom:
01 Compute carriers: a rich set of node types
ChatDev 2.0 is no longer limited to a single LLM call; it offers multiple types of compute nodes:
Agent node: handles task execution and supports Function Calling, Memory, and Thinking.
Human node: native Human-in-the-loop support, so human feedback and decisions can enter an otherwise closed automated flow at any time.
Python node: executes input code and shares the Workspace.
Logic node: includes Subgraph, Loop_counter, and more, supporting nested loops of arbitrary depth and logic reuse.
02 Control and communication: the dual role of edges
ChatDev 2.0 decouples control flow from data flow on edges, separating logical judgment from information transfer and avoiding the awkwardness of triggering execution just to pass information:
Control flow: decides who executes next. It supports complex conditional branching and precisely schedules subsequent nodes.
Data flow: decides how information is passed. We implemented fine-grained context governance, supporting cleaning, extraction, and filtering during transfer. To fully support multi-agent parallelism for efficiency, edges also support dynamic dispatch: at runtime, the output of a preceding node is automatically split and distributed in parallel to multiple downstream nodes.
03 An orchestration revolution: unlimited nesting and free topology
ChatDev 2.0 refuses to flatten multi-agent systems into linear pipelines. It breaks the topology limits of most workflow tools and hands freedom back to developers:
Full topology compatibility with native support for cycles: the system supports not only directed acyclic graphs but most complex graphs with cycles. That means you can freely build nonlinear flows with multi-round self-reflection, review-and-revise loops, and more, capturing the real complexity of human collaboration.
Unlimited nesting, unlimited scale: ChatDev 2.0 natively supports unlimited nesting. You can wrap a complex collaboration group of dozens of agents into an ordinary “node” and call it directly from other graphs. This fractal architecture makes multi-agent systems scalable — from a simple three-person team to a vast virtual software company, the complexity is entirely up to you, with no upper limit.
Standard templates plus code-level extension: something for beginners and geeks alike
A true productivity tool needs both a very low barrier to entry and no capability ceiling. ChatDev 2.0 delivers not just a tool but a complete methodology for multi-agent collaboration, so developers at any level can build multi-agent systems efficiently and on their own terms:
01 Beginners: ready out of the box, no blank canvas
Best practices: ChatDev 2.0 ships with templates for many kinds of multi-agent systems. Whether productivity tasks (data analysis, Deep Research, video generation), creative design (3D modeling, game development), or classic reproductions (ChatDev 1.0, MacNet), all can be loaded and run with one click.
Hand-holding guidance: the built-in illustrated tutorial walks you through building your first workflow, so even absolute beginners can get started in five minutes.
Illustrated tutorial: a guided build of a multi-agent system
02 Geeks: open, transparent, and built for tinkering
Code-level extension: the system is fully open source. Developers can customize as deeply as they would with LangGraph: write new node types, connect private LLM providers, or rebuild the Memory storage backend. Custom Edge Conditions and Payload Processors let you clean data precisely and control flow direction.
Hardcore docs: detailed GitHub developer documentation, from YAML structure explanations to provider integration guides, helps you master the architecture quickly and turn ChatDev 2.0 into your own foundation.
Practical examples: exploring what ChatDev 2.0 can do
ChatDev 2.0 makes it easy to build multi-agent systems with different capabilities, efficiently handling complex tasks such as 3D modeling and data analysis and helping users produce results fast.
01 3D modeling
02 Data analysis
03 Video generation
04 Deep Research
05 Game development
We sincerely welcome the community to build ChatDev 2.0 with us.
Whether you share a multi-agent workflow you built, reproduce and validate existing research, contribute a representative application case, submit feature suggestions or bug fixes, or contribute new node types, orchestration patterns, and underlying mechanisms, we encourage and look forward to developers and researchers participating through PRs and issues.
For community contributors, we will:
List contributors in GitHub Contributors, and include high-quality community solutions, workflow templates, and application cases in our official examples and recommended practices.
Offer consistently active open-source contributors priority access to in-person technical events.
Explore further research collaboration where the direction and conditions fit, including potential internships, joint system development, and academic output.
If you care about the expressiveness and scalability of multi-agent systems and how they land in real scenarios, ChatDev 2.0 looks forward to exploring with you.
We believe the future of swarm intelligence belongs to everyone. Come to ChatDev 2.0 now and build your own super agent team, block by block!
Scan the QR code to join the group, read the detailed developer docs, and talk directly with the core team.