loki-auto

Stateless sandboxed browser automation runtime for AI Agents

Loki-auto is an ultra-lightweight, zero-dependency browser web automation runtime designed for the AI Agent era. It allows LLM agents (such as Claude Desktop, Cursor, or custom agents) to safely execute local browser automation scripts directly inside your active browser tab. By replacing heavy, resource-intensive headless browsers (like Puppeteer or Playwright) with a sandboxed WebAssembly execution environment, Loki-auto brings episodic, transient browser automation to your local workflow with maximum performance and zero-trust security. --- ### Key Features 🚀 Bypassing Headless Overhead No need to spin up heavy browser instances. Loki-auto executes automation scripts directly inside the active browser context on-demand, saving CPU and RAM. 🔒 WebAssembly Sandbox Isolation Built on Rust and WebAssembly, Loki-auto compiles the Rhai scripting engine to WASM. Automation scripts run strictly inside this VM sandbox with zero access to standard browser JS APIs (like cookies or localStorage) and zero system file access. 🛡️ Zero-Trust Tab-Level Gating LLM agents cannot interact with any tab by default. Users retain absolute control and must explicitly authorize specific Tab IDs via the extension popup/options page. Once a tab is closed, its permission is immediately and automatically revoked. 🌐 Local-First & Zero Network Sandbox The WebAssembly execution sandbox has zero network access. All operations are local-first, preventing prompt- injection attacks from leaking sensitive page content or cookies to external servers. 🔌 Standard Model Context Protocol (MCP) Support Compatible with the Model Context Protocol (MCP) standard. Seamlessly integrates with Claude Desktop, Cursor, and any LLM orchestrators over secure WebSocket/stdio bridges. --- ### How It Works 1. Connect the open-source Loki MCP server to your LLM client. 2. Open any web page you want to automate. 3. Click the Loki-auto extension icon and toggle "Authorize Tab" for that specific page. 4. Your LLM agent can now perform precise actions (clicking, typing, data extraction) via sandboxed Rhai scripts on the whitelisted tab. --- ### Privacy & Security Commitment Loki-auto is open-source and built with a privacy-first philosophy. * We do NOT collect, store, or transmit any user data. * All script compilation and DOM interactions happen entirely locally on your machine. * The extension only runs on tabs you explicitly whitelist. Source Code: https://github.com/loki4agent/loki-auto Documentation: https://docs.loki4agent.com