LeonAI

KO | EN

gitlite — search

#javascript #hacktoberfest #typescript #rust #deep-learning #nodejs #windows #awesome-list #vue #pytorch #security #mcp #ios #open-source

(★ 16)

No description available.

LeonAI Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Leon AI - Lightweight  Development Tool

Leon AI is a lightweight, multi-user development platform designed to facilitate user development, based on the Claude Code CLI development tool. It integrates a React-based front-end workspace with a FastAPI back-end, supporting real-time AI assistance, code editing, previewing, and workspace terminal execution.

---

## 🏗️ Architecture Overview

The system follows a classic **Client-Server** architecture tailored for local development and real-time AI-assisted iteration:

### 1. Frontend (Vite + React + TypeScript)
The frontend serves as the primary developer cockpit. Key modules include:
*   **Workspace Explorer**: Multi-tenant file navigator displaying directory trees, supporting file/folder CRUD, uploads, and raw resource downloads. Now supports a **draggable, resizable sidebar** with width settings preserved in `localStorage`.
*   **Monaco Code Editor**: Web-based IDE powering syntax highlighting and code editing, integrated with a keyboard shortcut hook for instant saving (`Ctrl+S`).
*   **Multi-Device Canvas Preview**: A responsive design preview system that renders HTML and screen prototypes inside an iframe, with viewport switching (Desktop, Tablet, Mobile).
*   **AI Chat Assistant**: Streamed real-time chat with file attachment support (docx, pptx, images, code files) for prompt engineering and automated code writing.
*   **Terminal Panel**: Integrated command-line console running commands directly within the isolated workspace.
*   **Admin Dashboard**: CRUD interface for managing user accounts, passwords, and permissions.

### 2. Backend (FastAPI + SQLite + WebSockets)
The backend manages workspace sandboxing, user management, and coordinates the LLM agent tool-execution loop:
*   **`server/app.py`**: API Gateway exposing REST endpoints for user authentication, file administration, terminal shell execution, project exports, and the live Chat WebSockets `/ws/chat`.
*   **`server/db.py`**: SQLite database layer storing credentials, roles (`admin` / `user`), and creation timestamps.
*   **`server/chat/ws_handler.py`**: Coordinates WebSocket connections. Compiles active workspace details, builds prompt contexts, and runs the LLM processing stream.
*   **`server/chat/claude_client.py`**: Interacts with the Anthropic Claude API to stream text delta updates and invoke tool calls.
*   **`server/chat/tool_executor.py`**: Sandbox tool executor validating and running agent operations:
    *   `read_workspace_file`: Reads files inside user workspace or templates.
    *   `write_workspace_file`: Safely writes code into workspace directory files.
    *   `list_dir`: Recursively lists directories.
    *   `create_folder`: Automatically spawns folder structures.
    *   `run_terminal_command`: Executes terminal commands within the workspace environment, restricted against destructive operations.
*   **`server/chat/context_builder.py`**: Automatically parses and structures plain text, Word documents (`.docx`), and PowerPoint presentations (`.pptx`) using specialized parsers to feed context into the LLM.

---

## 📁 Directory Structure

```text
leonAI-development/
├── frontend/                     # React Vite frontend project
│   ├── src/
│   │   ├── App.tsx               # Main layout and client logic
│   │   ├── App.css               # Workspace styling & responsive rules
│   │   └── main.tsx              # React mounting root
│   ├── package.json              # Frontend npm dependencies
│   └── vite.config.ts            # Vite bundler configuration
│
├── server/                       # FastAPI python backend
│   ├── chat/                     # AI assistant and parsing module
│   │   ├── claude_client.py      # LLM API connection client
│   │   ├── context_builder.py    # Context assembly helper
│   │   ├── docx_parser.py        # Word document parser
│   │   ├── pptx_parser.py        # PowerPoint slide text parser
│   │   ├── tool_executor.py      # Agent sandbox tools
│   │   └── ws_handler.py         # WebSocket manager
│   ├── app.py                    # Main REST & WebSocket router
│   └── db.py                     # SQLite database abstraction
│
├── prompts/                      # System prompts & generation templates
│   ├── system-prompt.md          # Core AI guidelines & rules
│   ├── gen-product-charter.md    # Product charter prompt template
│   └── gen-prd.md                # PRD template
│
├── workspace/                    # Isolated multi-tenant workspace root
│   └── leon/
│       └── [username]/           # Isolated space for each user
│
├── start.py                      # Concurrently launches backend & frontend
├── start.bat                     # Windows bat launcher shortcut
├── requirements.txt              # Backend python packages
└── CLAUDE.md                     # File placement decision policies
```

---

## ⚙️ Setup & Installation

### Prerequisites
*   Node.js (v16+)
*   Python 3.8+
*   Anthropic API Key (placed in `.env`)

### 1. Configure Environment Variables
Create a `.env` file in the root directory:
```env
ANTHROPIC_API_KEY=your_claude_api_key_here
SERVER_HOST=127.0.0.1
SERVER_PORT=8900
FRONTEND_PORT=5173
```

### 2. Run the Application
The `start.py` script checks dependencies, sets up the workspace, and runs both servers concurrently:
```bash
python start.py
```
Or double-click `start.bat` on Windows. The application will launch automatically at `http://localhost:5173`.

---

## 🔐 Multi-Tenant Sandboxing

Leon AI implements sandboxing policies to prevent unauthorized file system access:
1.  **Isolated CWD Paths**: All workspace operations (`read`, `write`, `list`, `terminal`) resolve strictly within `workspace/leon/[username]/`.
2.  **Path Traversal Prevention**: Files paths are normalized and validated using `.resolve()`. Any path outside the user's workspace triggers a `403 Access Denied` error.
3.  **Command Filter**: Shell command terminal execution blocks destructive commands (e.g. `rmdir /s`, `taskkill`, `..`, etc.) to keep host systems secure.

LeonAI

LeonAI Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

LeonAI

LeonAI Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README