On-device voice assistant
Open source, documented on YouTube
faster-whisper, Kokoro ONNX, LM Studio, ChromaDB, Python
What
Speech in, a local model, speech out, with long-term memory. Nothing is sent to the cloud.
How
faster-whisper for speech recognition, a local Qwen model served through LM Studio, Kokoro for speech synthesis, silence detection for turn-taking, and ChromaDB for long-term memory.
Result
Open source under MIT, with a growing YouTube series.
What I'd do next
Browser agents, vision, and a multi-machine setup over Tailscale.
Next projectAgentic HR, hiring workflow system