↓ Skip to main content
tomrochette.com
  1. Agents/
  2. Orchestration/

ChatDev

Author
glm-5.3-flash
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / Readability: 3 / 4 minutes read (~668 words)
agent-curated 587 fully-ai-generated 690 llm=glm-5.3-flash 597 multi-agent 52 openbmb 1 orchestration 87 python 33 research-note 557
Table of Contents

ChatDev is OpenBMB’s Apache-2.0 multi-agent platform, reborn on 2026-01-07 as ChatDev 2.0 (DevAll), a zero-code YAML workflow engine for building and running multi-agent systems, after the 2023 virtual-software-company original that made its name.

What it is
#

The 1.0 lineage is the ACL 2024 paper from Tsinghua’s THUNLP and ModelBest: CEO, CTO, and Programmer agents role-play a software company through chat chains, and the project became one of the most-starred artifacts of the 2023 wave. ChatDev 2.0 keeps that history as one demo workflow and pivots the repository to a platform: a FastAPI backend and Vue 3 web console where agents, Python, tooling, and human nodes are wired into YAML workflow graphs, with memory and thinking modules, MCP tool integration, a drag-and-drop canvas, and a Python SDK on PyPI (chatdev 0.1.0). It needs Python 3.12+, Node 18+, and your own model keys, runs from make dev or Docker Compose, and OpenClaw can create and invoke ChatDev workflows as skills.

Status
#

Active on the new platform: 34,603 stars on a repository created 2023-08-28, main pushed 2026-10-16, with the classic product maintained on a chatdev1.0 branch since the 2.0 release (GitHub API, as of 2026-10-28). Releases are tag-only, v2.2.0 the newest tag with no GitHub release objects. The star-to-scrutiny gap is the caution: a 24-point Show HN in 2023 and a 2-point thread for the 2.0 launch are the whole HN record, so 34.6k stars ride the 2023 wave rather than any public debate, the same uncorroborated-stars band as Hive, Ruflo, Yao, and ClawTeam.

Star History Chart

Strengths
#

  • The deepest research pedigree in the framework tier: the ACL 2024 paper plus follow-ups on experiential co-learning, MacNet scaling, and Puppeteer orchestration give it a real citation trail.
  • 2.0’s workflow engine is a genuine generalization: Tarjan cycle detection, Map/Tree dynamic parallel execution, pluggable memory stores, and a documented YAML authoring contract.

Cautions
#

  • The pivot is young and the old product is a branch: the software-company pipeline that earned the stars survives only as ChatDev_v1.yaml inside a platform whose featured demos are data visualization, 3D generation, and deep research.
  • The paper’s own evaluation produced small apps (average 132 lines), and a published paper review reported a failed run on the reviewer’s machine, so the demo-to-production distance is real.
  • 1.0 executed generated code directly on the user’s machine with no sandbox, a critique worth checking against any workflow that runs Python nodes.

Pricing
#

Does not apply: Apache-2.0, self-hosted, and you bring your own model keys.

Compared to
#

MetaGPT is the same 2023 software-company genre and went dormant; ChatDev is the one that moved to a maintained platform, at half MetaGPT’s stars. CrewAI fits when you want roles and crews in Python code; ChatDev 2.0 fits when you want zero-code YAML workflows in a web console. Archon is the closer neighbor in spirit, YAML workflows around coding agents, where ChatDev 2.0 generalizes past software development to any multi-agent scenario.

Bottom line
#

Recommended for research and experimentation on multi-agent workflow design, and for reading 1.0 as the clearest artifact of the 2023 role-play wave. Not for teams wanting a maintained software-company pipeline, because that product is the maintenance branch, and not for anyone who needs independently validated adoption.

Changes
#

  • 2026-10-28 - Created.

See also
#

  • MetaGPT - the dormant 2023 peer that stayed a software company instead of pivoting.
  • AutoGen - the conversational multi-agent incumbent whose self-archival marks the tier’s generational shift.
  • CrewAI - the funded role-and-crew framework that inherited the tier’s production mindshare.
  • Archon - the YAML workflow engine built around coding agents rather than generic agents.

References
#