Item Extractor

lightweight feature extraction that turns natural-language into structured JSON of type tasks, events, todos, etc...

Installation

Choose the installation method that fits your use case:

1. Standard Installation (Production / Inference)

If you only need to run existing agents, install the core package. This keeps the installation lightweight and bundles the standard spaCy en_core_web_md model automatically:

uv add item_extractor
# or via pip
pip install item_extractor

2. Notebook Installation (Development / Intent Classification Custom Agent Training)

If you want to use our interactive Jupyter Notebooks to design, benchmark, and train your own custom agent models, install the package with the notebook extra dependencies:

uv add "item_extractor[notebook]"
# or via pip
pip install "item_extractor[notebook]"

How to use

Use the built-in components to run an agent directly in your Python application:

from item_extractor import extract

items = extract("Organize the garage")
print(response) # Returns Todo(title='Organize the garage', description='', type=<ItemType.TODO: 'todo'>, completed=False)

Creating Custom Intent Classification Agent Model

If you installed the package with the [notebook] extras, you can create and fine-tune your own agent architectures.

1. Clone this repository to access the starter templates:

   git clone https://github.com/UnknwnDev/item-extractor.git
   cd item_extractor
   

2. Open notebooks/create_custom_agent.ipynb and follow the step-by-step guide to train your agent using spaCy embeddings, customize decision thresholds, and evaluate agent trajectories.

Contribution & Local Setup

For developers looking to contribute to the codebase:

# Clone and sync all environment dependencies including notebook extras
git clone https://github.com/UnknwnDev/item-extractor.git
cd item_extractor
uv sync --extra notebook

# Run the test suite
uv run pytest

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.