Anthropic’s release of Claude Co-Work prompted Camel AI to open source their own multi-agent system, now called OpenWork, aiming to foster innovation and community collaboration in the AI agent space. OpenWork features a modular architecture, supports both proprietary and open-source models, and is designed for advanced multi-agent collaboration, positioning Camel AI as a leader in customizable agent frameworks.
Last week, Anthropic released Claude Co-Work, a significant development in the AI space that caused concern among startups working on similar products. One such startup, Camel AI, responded in a unique way. Instead of panicking or trying to compete directly with Anthropic, Camel AI made the bold decision to open source their own multi-agent system product, Agent AI, under the new name OpenWork. This move was announced via a tweet by Camel AI’s founder, Guhao, which quickly gained traction and widespread attention in the AI community.
Camel AI has a history of innovation in the agent framework space, having started as an academic project focused on training models for action-taking tasks during the early days of Llama 1. Over time, they evolved into a startup with the goal of building systems and frameworks for multi-agent architectures. Unlike more mainstream frameworks like LangChain or LlamaIndex, Camel AI’s focus has been on enabling research and experimentation with large-scale, customizable agent systems, rather than on mass adoption.
The OpenWork system is architected in three tiers: a React Electron desktop frontend, a FastAPI backend for orchestration and state management, and a workforce layer built on Camel AI’s multi-agent core. The system is designed to support both proprietary and open-source models, with a long-term vision of enabling local, on-device agent execution. OpenWork is already compatible with models from various providers, including Gemini, MiniMax, Zhipu AI, and Moonshot, and the team is committed to making it increasingly accessible for local deployment as open models continue to improve.
Technically, OpenWork employs advanced patterns for multi-agent collaboration, such as task decomposition, coordination via a directed acyclic graph (DAG), and parallel execution of subtasks. It comes with four specialized agents: a developer agent (for code execution and file system tasks), a browser agent (for information retrieval and web automation), a document agent (for writing and editing), and a multimodal agent (for handling images and audio). The system also features a custom browser automation toolkit built on Playwright APIs, enabling sophisticated navigation and interaction capabilities.
By open sourcing OpenWork under an Apache 2 license, Camel AI aims to foster innovation and community involvement in multi-agent systems. The team is not focused on selling to developers but rather on advancing the field and attracting talented agent builders for industry-specific enterprise solutions. The project has already seen significant interest and contributions from the open-source community, and Camel AI continues to actively develop and support the platform. This pivot not only benefits the broader AI community but also positions Camel AI as a leader in open, customizable agent architectures.