VIMA v1.0.0 — Initial Release
Voice Intelligent Machine Assistant — a fully local, microservices-based voice AI assistant that runs entirely on your machine (no API costs, no cloud dependency) with an optional Gemini Flash fast-path when you need lower latency.
What's included
Core voice pipeline
- Speech-to-text via
faster-whisper(Python FastAPI service) — supportstiny,base,small,medium,largemodels - LLM reasoning — defaults to Gemini Flash when online, silently falls back to local Ollama when offline
- Text-to-speech via
edge-tts(Python FastAPI service) - End-to-end latency: ~3–5s with Gemini Flash, ~15–17s fully local on CPU (i7 8th gen, no GPU)
Tools the assistant can run
| Tool | Example command |
|---|---|
| youtubeOpen | "play Believer" |
| youtubeSearch | "search rickroll on YouTube" |
| playMusic | "play some lofi on Spotify" |
| sendEmail | "email mom saying I'll be late" |
| sendMessage | "WhatsApp dad I'm on my way" |
| openManhwa | "read Solo Leveling" |
Known limitations / future improvements
py-server/core/whisper.pyhardcodesdevice="cpu"— change to env var for GPU support- WhatsApp via Baileys requires scanning a QR code on first run
- Spotify tool requires a Premium account for full playback control
Full Changelog: v1.0.0...v1.0.0