But really it’s a general purpose tool for hierarchical clustering of text data using LLMs, based on text facets you extract from your data.

Designed to run on local LLMs, and outputs to a single static html file that can stream the data as needed via static, compressed json chunks.

You can see outputs from this tool on ~400,000 conversations using Qwen 3 8B at https://www.phylliida.dev/modelwelfare/wildchat/

The state of the tree/which conversation you have open is saved in hash for easy sharing.

While the prompts are “privacy preserving” I’m just interested in this for data analysis of public data so that feature wasn’t very important to me and you can see the individual conversations as well.