README.md

---
title: Wllama
emoji: 🦙
colorFrom: green
colorTo: pink
sdk: static
pinned: true
short_description: Run GGUF directly on your browser!
license: mit
header: mini
custom_headers:
  cross-origin-embedder-policy: require-corp
  cross-origin-opener-policy: same-origin
  cross-origin-resource-policy: cross-origin
---

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

## Running locally / offline

Serve this folder with the COOP/COEP headers the Space uses (required for
multithreading + the model cache), then open the printed URL:

```bash
python3 serve.py            # http://localhost:8000
```

What's offline-ready after one-time downloads:

- **VAD + TTS**: bundled locally (`vad/`, `piper/`, `voices/`, `onnx/`) — always work offline.
- **LLM model (GGUF)**: download once from the *Manage models* page; it is cached in the browser.
- **ASR (Whisper)**: use the **Voice modules → Install** button on the *Manage models* page.
  It downloads the whisper model and runs the ONNX runtime from the local `asr/` folder,
  so speech-to-text works fully offline after install.

## Voice-to-voice mode

Click **🎙 Voice** in the top bar for a full-screen voice interface:

- Hands-free: speak → transcribed (whisper) → answered (LLM) → read aloud (piper TTS).
- A neon audio visualizer reacts to the assistant's speech.
- Only sentences that have already been spoken are shown on screen, so faster
  text generation stays hidden until it is actually said.
- On a new conversation the LLM invents a random personality for the assistant
  and speaks its greeting.
- Click any spoken sentence (or any assistant bubble in normal chat) to
  **regenerate** that reply.