AI-powered Typing Assistant with local (through Ollama) or external models (through PyLLMs)

This script runs in the background and listens for hotkeys, then uses a Large Language Model to fix or improve text. It's inspired by Andrej Karpathy's tweet about GPT's ability to correct minor typos, allowing for faster writing.

Features

• Uses Ollama or any external model supported by PyLLMs for text processing
• Supports multiple hotkeys for different text operations
• Can fix typos, improve writing style, and restructure notes

Setup

1. Set up Ollama (if using Ollama):

   • Install Ollama: https://github.com/ollama/ollama
   • Run: ollama run mistral:7b-instruct-v0.2-q4_K_S

2. Install dependencies:

   pip install pynput pyperclip httpx pyllms
   

3. Run the script:

   python key.py
   

Usage

Default Hotkeys:

• F8: Restructure and polish the entire note
• F9: Fix typos in the current line
• F10: Fix typos in the current selection
• F11: Improve writing style of the current selection

Note: On macOS, you may need to add the script (IDE/terminal) to both accessibility and input monitoring settings.

Customization

The script can be easily customized:

• Modify OLLAMA_CONFIG to change the Ollama model or settings
• Edit PROMPT_TEMPLATES to change how text is processed
• Adjust function_key_map to modify hotkey assignments or add new operations

To extend mappings:

1. Add a new prompt template to PROMPT_TEMPLATES:

   "new_operation": Template("Your prompt here $text"),

2. Add a new entry to function_key_map:

   "<key_code>": ("Key_Name", select_function, "new_operation"),

   Where <key_code> is the pynput key code, "Key_Name" is a descriptive name, select_function is either None or a function to select text, and "new_operation" is the key of your new prompt template.

Command-line Options

• Use -m or --model to specify a PyLLMs model instead of Ollama

Examples:

python key.py -m gpt-4o-mini

Notes

• The script is designed for macOS. Keyboard shortcuts may need adjustment for Linux or Windows.
• Ensure you have the necessary permissions for keyboard input monitoring.
• When using PyLLMs models, make sure you have the required API keys set up in your environment.