Run an AI coding agent inside a browser sandbox without setting up any server infrastructure.
Define YAML policies that restrict which files, processes, and ports an agent can access.
Review a full audit log of every file, network, and process action an agent performed.
Clone and push GitHub repositories directly from an agent running inside the browser.
| open-gitagent/clawless | glassflow/clickhouse-etl | salhanabil/cloakbrowser | |
|---|---|---|---|
| Stars | 462 | 474 | 479 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | easy | hard | easy |
| Complexity | 3/5 | 4/5 | 3/5 |
| Audience | developer | data | developer |
Figures from each repo's GitHub metadata at analysis time.
Needs an API key from Anthropic, OpenAI, or Google to run an actual agent.
ClawLess is a tool for running AI coding agents, called Claw Agents, directly inside a web browser instead of on a server. It uses a browser technology called WebContainers, which lets a full Node.js environment run as WebAssembly, giving the agent access to millions of npm packages without needing any backend infrastructure. Everything from writing code to running it and previewing the result happens inside an isolated sandbox in the browser, with no access to the user's actual computer. The project includes a built in code editor based on the same technology used in VS Code, along with a real terminal for running commands. It connects to GitHub so agents can clone and push repositories. A policy engine, written in YAML, lets a developer define exactly what an agent is allowed to do: which files it can touch, which processes it can start, and limits like maximum file size or number of turns. Every action the agent takes inside the sandbox, including file changes, network requests, and process activity, is recorded in a detailed audit log, with sensitive values like API keys automatically hidden. ClawLess supports agents built with multiple AI providers, including Anthropic, OpenAI, and Google, selected through configuration variables along with the relevant API key. It also has a plugin system for extending its behavior and a template system for quickly setting up new agent configurations. Under the hood it is built with Vite and TypeScript, using the WebContainer API for the sandboxed runtime, xterm.js for the terminal, and Monaco Editor for the code editing interface. To try it, a developer can clone the repository and run it locally with npm, or install it as a package called clawcontainer inside their own project. The README describes it as built on something called the GitAgent Standard and licensed under MIT.
A browser-based sandbox for running AI coding agents entirely client side, with a built-in editor, terminal, GitHub integration, and policy-based guardrails.
Mainly TypeScript. The stack also includes TypeScript, Vite, WebContainer API.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice, under the MIT license.
Setup difficulty is rated easy, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.