camel-ai/owl

OWL (Optimized Workforce Learning) is a Python framework for building systems where multiple AI agents collaborate to complete complex, real-world tasks automatically. The idea is that rather than using a single AI model to handle an entire task, you can assemble a team of specialized agents that each contribute their own capabilities, one browsing the web, another writing code, another searching documents, and they coordinate to reach a goal. The framework is built on top of CAMEL-AI and focuses on making this multi-agent coordination efficient and practical. Agents can be equipped with a wide range of tools, including web browsing via Playwright, web search, file writing, terminal access, and integrations with the Model Context Protocol (MCP), a standard for connecting AI assistants to external tools. The system supports many underlying language models including those from OpenAI, Google Gemini, and others. OWL achieved a score of 69.09 on the GAIA benchmark, a test of general AI assistant capabilities on realistic tasks, ranking first among open-source multi-agent frameworks at the time. The research behind it was accepted at NeurIPS 2025. The team has also released training datasets and model checkpoints, with training code forthcoming. A developer or researcher who wants to automate multi-step workflows, such as gathering information from the web, processing it, writing a report, and sending it, would use OWL as the orchestration layer for those agent pipelines. It is written in Python, open source, and includes a web-based user interface for running tasks interactively.