iBox - Agentic AI Terminal UI

An interactive terminal user interface (TUI) for interaction with Mistral AI models, featuring autonomous tool execution, real-time streaming responses, and multi-model consultation capabilities.

Features

• Interactive Chat Interface - Modern terminal UI built with Textual framework
• Agentic Tool Execution - AI autonomously executes tools to accomplish tasks:
  • File creation and editing
  • Command execution
  • Directory management
  • Web searching
  • Code generation (via Codestral)
  • Image analysis (via Pixtral)
• Real-time Streaming - Watch AI responses appear in real-time as they're generated
• Deep Think Mode - Consult multiple models simultaneously and synthesize responses
• Model Switching - Easily switch between Mistral Small and Large models
• Keyboard Shortcuts - Efficient keyboard-driven workflow
• Tool Extensibility - Easy to add new tools via JSON configuration

How to start

Prerequisites

• Python 3.8 or higher
• Mistral AI API key (get one here) Or there is a hard-coded one

Installation

1. Clone the repository:

git clone https://github.com/Scutoidzz/iBox
cd iBox

2. Install dependencies:

pip install -r requirements.txt

3. Set your API key:

# Windows PowerShell
$env:MISTRAL_API_KEY="your-api-key-here"

# macOS/Linux
export MISTRAL_API_KEY="your-api-key-here"

or use the hard-coded default

4. Run the application:

python ibox.py

Usage

Keyboard Shortcuts

Shortcut	Action
q	Quit the application
d	Toggle dark mode
ctrl+l	Clear chat history
ctrl+m	Switch to Mistral Small model
ctrl+n	Switch to Mistral Large model
ctrl+j	Select a working directory

Commands

• /deep <prompt> - Uses Deep Think mode, which:
  1. Consults 5 different Mistral models in parallel
  2. Returns responses from each model
  3. Synthesizes them into a unified answer using the largest model

Available Tools

The AI can use the following tools to accomplish tasks:

Tool	Description
create_file	Create new files with content
read_file	Read and display file contents
edit_file	Modify existing file contents
delete_file	Remove files
run_command	Execute shell/terminal commands
create_directory	Create new directories
delete_directory	Remove directories
list_files	List directory contents
list_directories	List subdirectories
download_file	Download files from URLs
analyze_image	Analyze images using vision model
code	Generate code using Codestral
web_search	Search the web for information
read_github_repo	Fetch content from GitHub repositories

Project Structure

iBox/
├── ibox.py              # Main TUI application
├── tool_handler.py      # Tool execution engine
├── tools.json           # Tool definitions and schemas
├── code.py              # Code generation module
├── image_analyze.py     # Image analysis module
├── requirements.txt     # Python dependencies
├── pyproject.toml       # Project configuration
├── README.md            # This file
└── tests/
    └── test_tool_handler.py  # Tool handler tests

Architecture

Core Components

1. ibox.py - The main application using Textual framework

   • ChatMessage - Widget for displaying chat messages
   • SendingBar - Visual indicator for processing
   • TUIApp - Main application class handling chat logic

2. tool_handler.py - Tool execution system

   • Loads tools from tools.json
   • Executes tools synchronously
   • Handles tool result formatting

3. tools.json - Tool schema definitions

   • OpenAI function calling format compatible
   • Describes each tool's parameters and requirements

Configuration

Changing Models

The application supports multiple Mistral models:

• mistral-small-latest - Faster, suitable for simple tasks
• mistral-large-latest - More capable, better reasoning
• codestral-latest - Specialized for code generation
• pixtral-large - Vision model for image analysis

Switch models in-app with ctrl+m and ctrl+n, or modify the MODEL variable in code.

Dependencies

• mistralai - Mistral AI Python SDK
• textual - Python TUI framework
• rich - Rich terminal formatting and markdown support

Contributing

Contributions are welcome! Areas for improvement:

• Additional tool implementations
• UI/UX enhancements
• Performance optimizations
• Test coverage

License

MIT License - See LICENSE file for details