0.5-1.1 second round-trip latency with Whisper STT + Kokoro TTS + your local LLM. Needs 2-4GB extra VRAM on top of your model. Setup with Open WebUI and standalone options.