BMO - AI voice assistant

Build a fully local, voice-driven AI assistant with a model-agnostic brain - swappable between local Ollama (Llama 3.2) and hosted Claude/GPT-5 mini, each deciding natively which tool to call. BMO features wake-word detection, GPU-accelerated speech-to-text, streaming text-to-speech, and a stationary desktop companion overlay for natural interaction. Equipped with tools including YouTube music playback via yt-dlp, web search, live weather, timers, and system control, plus a RAG pipeline backed by PostgreSQL + pgvector with cross-encoder reranking for accurate responses.