0.3.6: dashboard in Traditional Chinese

0.3.5: check every reply phrase for leaked reasoning and keep it out of the chat history

0.3.4: speak English replies with a native voice (Piper en_US-lessac-low); Chinese stays on Matcha

0.3.3: transcribe sentence pieces while the user talks; score voice alongside speech-to-text

0.3.2: default fusion weights text 0.9 / face 0.5 / voice 0.3 (tuned on the robot)

0.3.1: SenseVoice speech-to-text by default (accuracy), never speak leaked model reasoning, 3 s reply deadline

0.3.0: talks back - streaming zh+en speech-to-text, free OpenRouter replies, local Matcha voice, prepared-line fallback

0.2.1: retry model download/loading instead of idling after a network error

0.2.0: one fused emotion and one move per spoken sentence; head follows the face; clearer dashboard

0.1.2: survive short daemon-link drops while models load; lighter ONNX Runtime threading

Bump to 0.1.1

Declare opencv-python-headless and pydantic (reachy-mini 1.11 no longer ships OpenCV)

Reachy Mini multimodal emotion app

initial commit