Timmy provides a browser catalogue for GGUF models and a local library of downloaded models. It connects that library to Pi's model list.
Behavior
Models come from Hugging Face and run through llama-server. Timmy handles downloads and starts a server for the selected model. One model is deployed at a time.
The catalogue and inference run on the local machine. Available memory and the model's compatibility with llama.cpp determine what can run.
Setup
Timmy requires Node.js, Pi, and llama-server. The repository README covers installation and the local model workflow.