Air Assistant

Air Assistant is a versatile application designed to enhance your productivity by allowing you to ask and interact with your screen content using AI and a suite of powerful tools.

Demo Video

Features

• Screen Interaction: Ask questions about anything displayed on your screen.
• AI Automation: Leverage AI to automate tasks and streamline your workflow.
• MCP Tools: Access a variety of Model Context Protocols (MCP) tools for extended functionality.

Tech Stack

• Gemini
• OmniParser
• MCP
• OpenCV
• Mediapipe
• PyQt5

Prerequisites

• Python 3.12
• uv
• Hugging Face account and CLI setup

Local Setup and Execution

Follow these steps to get Air Assistant running on your local machine:

1. Install uv (if not already installed):

   pip install uv

2. Clone the Repository (if you haven't already):

   git clone https://github.com/boedegoat/air-assistant.git
   cd air-assistant

3. Install Dependencies:

   uv sync

4. Configure Environment Variables:

   • Create a .env file in the project root by copying .env.example:

     cp .env.example .env

   • Update the .env file with your specific configurations (e.g., API keys).

5. Set up OmniParser:

   • Ensure your Hugging Face CLI is configured (huggingface-cli login).
   • Download OmniParser model weights:

     cd omniparser
     # Download model checkpoints to OmniParser/weights/
     for f in icon_detect/{train_args.yaml,model.pt,model.yaml} icon_caption/{config.json,generation_config.json,model.safetensors}; do huggingface-cli download microsoft/OmniParser-v2.0 "$f" --local-dir weights; done
     mv weights/icon_caption weights/icon_caption_florence
     cd ..

   • In the omniparser/omnitool/client directory, create a specific .env file for the OmniParser client:

     cp omniparser/omnitool/client/.env.example omniparser/omnitool/client/.env

     (Adjust values in omniparser/omnitool/client/.env)

6. Configure MCP Tools:

   • Create servers_config.json from servers_config.example.json in the project root:

     cp servers_config.example.json servers_config.json

   • For the filesystems server within servers_config.json, adjust the path to your local filesystem as needed.

7. Run the Application Components (in separate terminals):

   • OmniParser Server:

     uv run omniparser/omnitool/server/omniparserserver.py

   • Main AI Client (defaults to screen mode):

     uv run main.py

   • Virtual Mouse (optional):

     uv run virtual_mouse/virtual_mouse.py

Future Plans

Key areas for future development include:

• Accuracy Enhancement: Continuously improving the precision and reliability of AI features.
• User Interface (UI) Improvements: Developing a more intuitive and user-friendly graphical interface.
• Expanded MCP Toolset: Integrating additional MCP tools to broaden the application's capabilities.
• Containerization: Packaging the application (e.g., using Docker) for easier deployment and scalability.