Wide-Moat's Open Computer Use: LLM Sandbox with Docker

What changed

Wide-Moat's Open Computer Use project has seen its latest release, version v0.11.0.0-rc.1. This open-source, self-hosted MCP server is designed to provide Large Language Models (LLMs) with a comprehensive computing environment. The system offers managed Docker workspaces that include a live browser with CDP streaming, an interactive terminal, and the capability to execute code. It also supports document creation skills for formats like Word, Excel, and PowerPoint, and can run autonomous sub-agents, such as Claude Code.

The project is built for production multi-user deployments, with each chat session isolated in its own Docker container to prevent data leakage between users. It is designed to be pluggable into any MCP-compatible client, including Open WebUI, Claude Desktop, and n8n, ensuring backend consistency across different frontends.

Key differentiators highlighted in the project's documentation include its self-hosted nature, compatibility with any OpenAI-compatible LLM (unlike some proprietary solutions), full Linux sandbox for code execution, and live browser streaming via CDP. It also features a skills system with over 13 built-in and numerous example skills for tasks ranging from document generation to web testing and AI-driven design.

While the project is actively maintained with fixes and security updates, the pace of new feature development is expected to slow as the Wide-Moat team focuses on their enterprise AI platform, Wide Moat. The project's license, FSL-1.1-Apache-2.0, ensures that users can continue to use, fork, and redistribute the software, with releases converting to Apache-2.0 two years after publication.

Why it matters for builders

Open Computer Use provides developers with a powerful sandbox for LLMs, enabling them to perform complex tasks that require direct interaction with a computing environment. Builders can leverage this to create more sophisticated AI agents capable of writing and executing code, automating web interactions, and generating professional documents. The self-hosted and open-source nature of the project offers flexibility and control, allowing integration into existing workflows and custom applications without vendor lock-in.

Practical impact

Developers can try out Open Computer Use via a free online demo at lab.widemoat.ai, which includes Open WebUI and bundled models, accessible with GitHub or Google sign-in. For full control, self-hosting is available. The quick start guide involves cloning the repository, setting an OpenAI API key in the .env file, and running docker compose up --build for the server and docker compose -f docker-compose.webui.yml up --build for Open WebUI. It's crucial to configure Model Settings in Open WebUI to Function Calling: Native and Stream Chat Response: On for the tools to function correctly.

Builders can explore the extensive list of tools available within the sandbox, including Python, Node.js, Java, LibreOffice, Playwright, and various AI and media libraries. The project also offers a rich set of 13 built-in skills and 14 example skills for tasks like creating presentations, editing documents, automating web scraping, and delegating tasks to sub-agents.

Caveats and source limits

The source indicates that while the Open Computer Use repository will continue to be maintained for fixes and security updates, the pace of new feature development will be slower due to the team's focus on the enterprise Wide Moat platform. The project's primary focus is on providing a stable, self-hosted environment rather than rapid innovation. The online demo at lab.widemoat.ai may have periods of offline status due to ongoing transformations. The source does not provide specific performance benchmarks or independent verification of its multi-user deployment capabilities beyond stating it has been tested with 1,000+ MAU.