Run, observe, and control AI agents entirely in the browser — no backend required. ClawLess provides a full sandboxed Node.js environment via WebContainers (WASM) with built-in editor, terminal, policy engine, and audit logging.

See It In Action

ClawLess runs a full Node.js runtime in the browser via WebContainers — that means access to 3.4 million+ npm packages. In this example, the agent installs pptxgenjs, generates a polished PowerPoint file with charts, images, and styled layouts, and saves it to the virtual filesystem — all without a server. The agent even learns and crystallizes reusable skills for future tasks.

The agent runs inside a WebContainer with full virtual filesystem access — reading, writing, and executing files in an isolated WASM runtime. No server, no backend, no access to the host system. Everything from code generation to live preview happens within the sandboxed browser environment, completely isolated from your machine.

Every action inside the container is tracked end-to-end: process lifecycle, file I/O, network requests/responses, environment configuration, and policy enforcement. Sensitive headers like API keys are automatically masked. Filter by source, level, or event type — and download the full audit log for compliance and debugging.

Control agent behavior with a built-in policy system. Define file access rules, allowed processes, port bindings, and runtime limits like max file size, max processes, max turns, and timeout — all in a simple YAML format. Policies are enforced at the container level, so agents cannot bypass them. Apply or reset policies on the fly without restarting the container.

Key Features

• WebContainer-powered sandboxed runtime (WASM) — full OS-level isolation in the browser
• Monaco Editor with multi-file tabs — rich editing experience out of the box
• xterm.js terminal with full PTY support — real terminal sessions, not a toy console
• GitHub integration — clone and push repositories via the GitHub API
• YAML-based policy engine with glob patterns — declarative guardrails for agent behavior
• Complete audit logging — process, file, network, and git events captured end-to-end
• Plugin system with lifecycle hooks — extend and customize every stage of execution
• Template system for agent configurations — bootstrap agents from reusable presets
• Network interception — intercepts both browser fetch and Node.js http calls
• Multi-provider AI support — Anthropic, OpenAI, and Google out of the box

Quick Start

# Run locally
git clone https://github.com/open-gitagent/clawless.git
cd clawless
npm install
npm run dev

# Install as a dependency
npm install clawcontainer

SDK Usage

import { ClawContainer } from 'clawcontainer';

const cc = new ClawContainer('#app', {
  template: 'gitclaw',
  env: { ANTHROPIC_API_KEY: 'sk-...' }
});

await cc.start();
cc.on('ready', () => console.log('Container ready!'));

Architecture

| Component | Role | |---|---| | ClawContainer | SDK facade — the single entry point for consumers | | ContainerManager | WebContainer orchestration and lifecycle | | PolicyEngine | YAML-based guardrails enforcing file, process, and network rules | | AuditLog | Complete event trail for every action inside the container | | GitService | GitHub API integration (clone, commit, push) | | PluginManager | Lifecycle hooks for extending container behavior | | UIManager | Monaco Editor, xterm.js terminal, and tab management |

Tech Stack

• Vite + TypeScript — fast builds, type-safe codebase
• WebContainer API — browser-native OS environment
• xterm.js — full-featured terminal emulator
• Monaco Editor — the editor behind VS Code

Configuration

ClawLess is configured through environment variables passed to the ClawContainer constructor:

| Variable | Purpose | |---|---| | ANTHROPIC_API_KEY | Anthropic API key | | OPENAI_API_KEY | OpenAI API key | | GOOGLE_AI_API_KEY | Google AI API key | | CLAWLESS_MODEL | Model selection (e.g. claude-sonnet-4-20250514, gpt-4o) |

All runtime state is persisted to localStorage under the clawchef_ prefix, so sessions survive page reloads.

Links

Supported Providers

| Provider | Models | |---|---| | Anthropic | Claude Sonnet, Claude Opus, Claude Haiku | | OpenAI | GPT-4o, GPT-4, GPT-3.5 | | Google | Gemini Pro, Gemini Flash |

Roadmap

• [ ] Custom agent template marketplace
• [ ] Multi-agent orchestration
• [ ] Persistent filesystem across sessions
• [ ] Cloud deployment support
• [ ] Built-in agent debugging tools

Community

• GitHub Discussions — ask questions, share ideas
• Issues — report bugs, request features
• Contributing Guide — how to contribute

Links

Documentation | Contributing | License | GitAgent Standard