Detoxify

Package

Locally run toxic comment classifier models for Python

Price: Free, open source
Access: None

Repo

About

Detoxify is a Python package with trained models that score comments for toxicity, including threats, insults and identity attacks. It runs locally on Transformers and PyTorch Lightning. Original, unbiased and multilingual models are provided, the last covering Turkish among others.

What you can do with it

Score user comments for toxicity without sending them to a vendor
Flag abusive messages for human moderators to review
Screen comments in several languages with the multilingual model

Get started

Install the package with pip
Load a model and score your text

Example

pip install detoxify
python -c "from detoxify import Detoxify; print(Detoxify('original').predict('example text'))"

Details

Hosting: Runs locally
Available in: Worldwide
Official SDKs: Python
MCP server: None