OWL

OWL is a Python framework built on CAMEL-AI for developers who want multiple AI agents to work together on tasks. It combines research, file analysis, and actions in a browser or terminal within one task workflow.

• Web research and browser tasks: Agents can search Google, DuckDuckGo, Wikipedia, or SearxNG and use Playwright to interact with websites, fill forms, and download files.
• Documents and media: Extract text or Markdown from Word, Excel, PDF, and PowerPoint files, and analyze images and videos with a multimodal model; audio processing requires the OpenAI API.
• Code and file actions: Agents can write and execute Python, run terminal commands, and save files as part of a task.
• Tool connections: MCP support connects agents to local or remote tools and data sources, alongside integrations for GitHub, Notion, arXiv, and other services.
• Local web interface: A Gradio interface provides agent chat, model selection, and a history of tasks and results.

OWL runs on macOS, Windows, and Linux, with Docker also supported. It works with Ollama, OpenAI-compatible APIs, OpenAI, Claude, Qwen, DeepSeek, Gemini, and Azure OpenAI. The framework and web interface run locally; configured cloud models receive data through their APIs, and web search needs internet access. Models must support tool calling, and visual tasks require multimodal capabilities. The maintainers recommend OpenAI models for complex tasks and report lower performance with some alternatives.