always-on-ai-assistant

Why we included this project

This repo is less a finished tool than a working blueprint for a voice-driven engineering assistant. It binds a DeepSeek-V3 brain to RealtimeSTT, so you can say a request aloud and have it turned into a runnable Typer command instead of typing it yourself. The demo shows the full loop in action: the assistant writes what it comes up with to a scratchpad file, and a mode flag decides whether that command actually runs. The config file names each part of the pipeline explicitly, which makes this a handy reference for anyone wiring a speech interface to an agent-style command executor. Just don't expect plug-and-play: it is a starter pattern, so you supply your own DeepSeek and ElevenLabs API keys, and the base assistant still runs on local Ollama models.