Echo Home

Cloud smart speakers ship house audio to third-party servers, answer from rigid command menus, and can't use your own cameras, files, or tools.

The house only carries sound. A custom "Echo" wake word (openWakeWord) and Whisper speech-to-text in each room feed one persistent Muse (Muse Spark) agent session that reasons with its own tools — shell, files, web search, house sensors — with no hardcoded menus. Kokoro text-to-speech answers on the speakers of the room you are in, picked per turn by loudest mic, Frigate person detection and three ESP32 Wi-Fi CSI presence nodes, so the conversation follows you from room to room. An owner voiceprint gate decides who may speak, and privacy and game modes switch by voice. Frigate watches seven cameras — five RTSP plus two custom ESP32 nodes at about 13 frames per second — on dual CPU detectors, recording to the home server. Live-proven end to end in October 2026.