uezo/aiavatarstackchan

AIAvatarStackChan is a framework for turning a StackChan (a small physical robot built on the M5Stack hardware platform) into a voice-interactive AI companion. The robot can hold conversations, display facial expressions, move its head, take photos, and respond to being touched. The project combines firmware that runs on the robot itself with a Python server that handles the AI processing. The setup has two parts. A Python server runs on a computer or server and handles speech-to-text, language model inference, and text-to-speech. These components are all swappable, so you can connect different AI services as they evolve. The firmware side runs on the StackChan hardware and handles audio capture, display, servo control, and communication with the server over WebSocket. You configure Wi-Fi credentials and the server address in a JSON file placed on the robot's SD card. Once running, the robot listens for your voice and sends audio to the server for processing. It shows a cartoon face with multiple expressions, automatic blinking, and lip sync animation that follows the speech output. If you say something like "look at this", the robot snaps a photo with its camera and sends the image to the server so the AI can respond to what it sees. Petting the top of the device triggers a reaction. You can also configure push-to-talk for noisier environments. The framework includes a hooks system so you can add custom behavior at key moments, such as when speech is detected, when the server accepts input, when a tool call runs, or when the robot finishes responding. An optional integration with an agent system called OpenClaw lets the robot take real-world actions beyond just conversation. English and Japanese configurations are both included. Japanese voice output requires the VOICEVOX speech synthesis engine running locally alongside the Python server.