A voice assistant that never phones home

What it does june is a command-line voice assistant that runs entirely on your own hardware. It listens through your microphone, transcribes speech with OpenAI’s Whisper via Hugging Face Transformers, generates a response using a local Ollama model, and speaks back through Coqui TTS. You can also fall back to plain text on either side if you prefer typing or reading.

The interesting bit There is no wake word: the tool simply starts recording when it hears sound and stops after three seconds of silence, which feels more like a conversation and less like addressing a smart speaker. It also supports voice cloning through Coqui’s XTTS v2 models, so you can make your assistant sound like someone specific with about a minute of sample audio.

Key highlights

• Runs fully offline; no audio or text leaves the machine by default
• Four interaction modes, from text-only to full voice-in/voice-out
• Configurable via a JSON file that overrides defaults partially rather than replacing them entirely
• Supports remote Ollama instances via environment variable if you want to offload the LLM
• Cross-platform, though it requires PortAudio and various system development headers

Caveats

• Requires manual installation of system dependencies like PortAudio and Python development headers before it will run
• The default Ollama model must be pulled separately; the tool does not handle model downloads itself
• Voice cloning and multilingual TTS require switching to specific Coqui models in the configuration file

Verdict Try it if you want a fully local voice assistant and are comfortable wrangling Python audio dependencies. Avoid it if you expect turnkey installation or cloud-synced history across devices.