Modular Local AI Engine
For maximum privacy, Zaya includes a split modular pipeline that runs entirely on-device without any cloud internet calls:
Microphone → WebRTC VAD → faster-whisper → Local LLM (Ollama) → Piper ONNX → Speaker
Setup Steps
- Install local speech dependencies:
.\.venv\Scripts\python.exe -m pip install -r requirements-local-speech.txt
- Install Ollama and pull your model:
ollama run qwen2.5:14b
- In Zaya Settings > AI ENGINE, select LOCAL PIPELINE, choose Ollama, and supply your Piper ONNX model path.