# Sembleverse

Sembleverse exports the current Semble link index and builds a local map of related links.

Both Python files are self-contained `uv` scripts. You'll also need `llama-embedding` from llama.cpp.

## Run

```sh
uv run --script export_semble.py
uv run --script embed_semble.py
```

Open `explore.html` in a browser after both commands finish.

## Embedding Model

- The map uses `ggml-org/embeddinggemma-300M-qat-q4_0-GGUF`, which is the public QAT Q4_0 build of EmbeddingGemma.
- Each link gets a normalized 768 dimensional vector.
- The configured prompt is `task: clustering | query:` because the same vectors support the topic map and related link search.

## License

Sembleverse is available under the [MIT License](LICENSE).