Offline voice assistant

I wanted a voice assistant with no cloud, no subscription and no audio leaving the house.

1. 01A thin handheld records audio and plays replies. Speech-to-text, text-to-speech and the LLM are served over HTTP from a 4 GB Pi.
2. 02Whisper for speech recognition and Piper for synthesis over the Wyoming protocol, with Ollama running the model.
3. 0314 bugs root-caused and logged, including out-of-memory reboots that pushed TTS off the handheld.

• The day-one build worked end to end.
• It became the origin of Hermes Agent.

• →Revive it on the current mesh network.