ClawLess
Browser AI agent runtime with an isolated Node.js sandbox, policy controls and audit logs. Uses cloud models from Anthropic, OpenAI and Google. MIT licensed.
ClawLess runs Claw AI agents in a browser sandbox, giving developers a place to execute and inspect agent tasks without a backend server. WebContainers provide a Node.js environment with a virtual filesystem that agents can read, write and execute files in, without access to the host system. The runtime runs locally. Model requests go to cloud services from Anthropic, OpenAI or Google using API keys, so this isn't fully offline AI.
The workspace includes Monaco Editor with multiple file tabs, an xterm.js terminal and live previews for apps an agent builds. Agents can use npm packages for tasks beyond code generation, such as creating PowerPoint files with pptxgenjs. GitHub integration supports cloning, committing and pushing repositories through the GitHub API.
A YAML policy engine controls file access, allowed processes, network behavior and execution limits. It enforces those rules at the container level, and users can change policies without restarting the container. Audit logs record agent actions alongside file operations, process activity, network traffic and Git events. Users can filter and download the logs; sensitive headers such as API keys are masked.
ClawLess is open source under the MIT license and builds on the GitAgent Standard. Its SDK lets developers embed the runtime in their own applications, while reusable agent templates and plugin hooks support custom workflows. Supported models include Claude Sonnet, Opus and Haiku, OpenAI GPT-4o, and Google Gemini Pro and Flash.
