lingliren731108-arch/leonai

Leon AI is a self-hosted development platform that wraps an AI assistant into a browser-based workspace where multiple users can write, edit, and run code with AI help. It combines a React frontend with a Python FastAPI backend, connected over WebSockets so AI responses stream in real time rather than appearing all at once after a delay. The browser interface includes a file explorer, a code editor with syntax highlighting and keyboard shortcuts for saving, a live preview panel that can render HTML pages at desktop, tablet, and mobile viewport sizes, an integrated terminal, and a chat panel where you can ask the AI to write or modify code. You can attach documents including Word files, PowerPoint presentations, and images to your chat messages, and the backend parses those files to extract text and feed it as context to the AI. Each user gets their own isolated directory on the server. File reads, writes, directory listings, and terminal commands are all restricted to that user's folder so one user cannot access another's files. The terminal blocks a set of destructive commands to reduce risk on the host system. The backend connects to the Anthropic Claude API to handle the AI assistant. When you send a message, the backend compiles your current workspace context, sends it to Claude, and streams the response back. Claude can also invoke tools: reading and writing workspace files, listing directories, creating folders, and running terminal commands inside your sandbox. An admin dashboard lets a user with admin privileges create and manage accounts. Setup requires Node.js, Python 3.8 or newer, and an Anthropic API key placed in a configuration file. A single startup script launches both the frontend and backend servers at the same time. The project also ships with prompt template files for generating product charters and requirement documents through the AI chat.