Skip to content

Introduction

LCXL Remote Desk is an AI-native, open-source, high-performance remote desktop. It treats AI as a first-class control plane alongside the browser: beyond browser-based remote control, it ships a built-in diagnostic AI agent that reads device state and can propose owner-confirmed commands, while exposing only the read capabilities to external AI assistants via a read-only MCP server.

The backend is written in Rust (Actix-Web); the frontend with React + Vite + Tailwind CSS.

Disclaimer

This project is currently in the early development stage. The codebase may be unstable, contain unfixed bugs, or have incomplete features.

Security Warning: Remote desktop technology involves deep access to computer systems. Ensure your network environment is secure when using this project. The author(s) shall not be held liable for any damages arising from the use of this project.

Why AI-Native?

The AI layer is security-first and model-agnostic (OpenAI-compatible and Anthropic APIs). The design rests on a few invariants:

  • The server is the sole authority on permissions — control planes can never self-report their identity, scope, or risk.
  • The model defaults to suggesting rather than executing — high-risk actions require explicit, server-mediated confirmation.
  • Data is strictly redacted before transmission, failing closed — redaction failure blocks the request before the model is ever called.
  • Every call is audited (metadata only) and API keys stay server-side.

See the AI Security Model for the full picture.

Key Features

  • AI Diagnostics — ask troubleshooting questions in plain language; the system collects and redacts read-only state for analysis. On the owner's own device, the agent may also request a supported-shell command, which remains blocked until the owner explicitly confirms that exact command.
  • Read-Only MCP Server--startup-mode mcp-stdio exposes a static whitelist of read-only tools to local AI assistants, with no execution or write permissions.
  • High-Performance Streaming — WebRTC with AV1 / H.264 / VP8 / VP9 encoding and Opus audio.
  • Remote Terminal — a built-in xterm.js terminal supporting full shell interactions.
  • File Management — uploads, downloads, deletions, and a recycle-bin mechanism.
  • Clipboard Sync — bidirectional text clipboard synchronization.
  • Remote Whiteboard — draw and annotate on the remote screen (requires tauri-app).
  • Privacy Screen — lock the local display and input during remote operations (requires tauri-app).
  • System Audio — captures and synchronizes remote audio playback.
  • Multi-language — UI and documentation available in English and Chinese.

Where to Next?

License

This project is licensed under the Apache-2.0 License.