ChatDev is an open-source, Python-based framework that simulates a virtual software company where multiple intelligent AI agents with distinct roles (such as CEO, CTO, programmer, tester, reviewer, and art designer) collaborate via chat-based communication to develop software. It leverages large language models (LLMs) to enable these agents to coordinate tasks across software development phases including design, coding, testing, and documentation. The framework is highly customizable and extendable, designed to make multi-agent collaboration accessible and practical for studying and creating complex software projects. Additionally, ChatDev incorporates mechanisms to reduce hallucinations through communicative patterns and supports incremental development and dockerized deployments for safety and scalability.
Key Features:
Multi-agent organizational structure with role-specific AI agents collaborating through chat chains for task breakdown and coordination.
Communicative dehallucination method to improve accuracy by requesting detailed information before responses.
Supports incremental development allowing agents to build upon existing codebases.
Docker support enables safe, isolated execution environments; also supports Git integration for version control.
Use Cases:
Automate end-to-end software development workflows by simulating a team of specialized AI developers.
Study and experiment with collective intelligence and multi-agent collaboration strategies in software engineering.
Create customized software solutions with minimal manual coding through natural language task descriptions.
Technical Specifications:
Implemented in Python (requires Python 3.9+), installable via pip with dependency management.
Configurable via JSON files defining organizational roles, phases, and software development processes.
Licensed under Apache 2.0 for source code and CC BY-NC 4.0 for datasets, emphasizing research use and non-commercial deployment.