Installation:
brew install whisper-cpp
Configuration:
Add to your moltis.toml: [voice.stt] provider = "whisper-cli" [voice.stt.whisper_cli] model_path = "~/.moltis/models/ggml-base.en.bin" Download models from Hugging Face.

See sherpa-onnx documentation for installation.

Configuration:
Add to your moltis.toml: [voice.stt] provider = "sherpa-onnx" [voice.stt.sherpa_onnx] model_dir = "~/.moltis/models/sherpa-onnx"

Installation:
pip install piper-tts Or download from GitHub releases.

Download a voice model:
Browse voices at piper-samples and download the .onnx file.

Configuration:
Add to your moltis.toml: [voice.tts] provider = "piper" [voice.tts.piper] model_path = "~/.moltis/models/en_US-lessac-medium.onnx"

Installation:
pip install TTS
Start the server:
tts-server --model_name tts_models/en/ljspeech/tacotron2-DDC Or use Docker:
docker run -p 5002:5002 ghcr.io/coqui-ai/tts
Configuration:
Add to your moltis.toml: [voice.tts] provider = "coqui" [voice.tts.coqui] endpoint = "http://localhost:5002"
Browse available models: coqui-ai/TTS

Requirements: Linux, Python 3.10+, CUDA GPU with 10GB+ VRAM

Setup:
1. Install vLLM: pip install "vllm[audio]"
2. Start the server:
vllm serve mistralai/Voxtral-Mini-3B-2507 \ --tokenizer_mode mistral \ --config_format mistral \ --load_format mistral 3. The model downloads automatically on first run (~3GB)

Configuration:
Add to your moltis.toml: [voice.stt] provider = "voxtral-local" [voice.stt.voxtral_local] endpoint = "http://localhost:8000"
Languages: English, French, German, Spanish, Portuguese, Italian, Dutch, Polish, Swedish, Norwegian, Danish, Finnish, Arabic

Run any OpenAI-compatible Whisper server locally. Popular choices:

faster-whisper-server (recommended):
pip install faster-whisper-server faster-whisper-server --model Systran/faster-whisper-large-v3 --port 8080
whisper.cpp server:
whisper-server --model ggml-large-v3.bin --port 8080
Configuration:
Add to your moltis.toml: [voice.stt] provider = "whisper-local" [voice.stt.whisper_local] endpoint = "http://localhost:8080"