XiaoZhi AI Voice Chatbot on LILYGO T-Circle S3
Run the viral XiaoZhi AI voice chatbot on the LILYGO T-Circle S3's tiny 0.75" round display — featuring offline wake-word detection, streaming voice interaction with Qwen/DeepSeek LLMs, and MCP protocol for smart home control.
Overview
XiaoZhi (小智) is an open-source ESP32-based AI voice chatbot with over 24,000 GitHub stars. It pairs offline wake-word detection with streaming cloud ASR + LLM + TTS to deliver fluid voice conversations. The LILYGO T-Circle S3 is one of the officially supported hardware targets, using the tiny round display for emoji expressions and status feedback while the device listens and responds.
Key Features
- Offline voice wake-up via ESP-SR — no cloud connection needed for activation
- Streaming voice interaction — speak naturally; responses stream as audio in real time
- Multi-LLM support — Qwen, DeepSeek, Doubao, and others
- MCP protocol — device-side MCP for controlling GPIO, LEDs, servos, and speakers; cloud-side MCP for smart home control, knowledge search, email, and PC desktop control
- Speaker recognition — identifies who is speaking using the 3D Speaker model
- Display support — shows emoji expressions and conversation content on the LCD
- Multi-language — Chinese, English, and Japanese voice recognition
T-Circle S3 Setup
The T-Circle S3's 0.75" round display shows chat status and emoji feedback. The onboard microphone handles voice input and the speaker handles TTS output. Firmware is flashed via the standard ESP32 Arduino/IDF pipeline. Customizable wake words, fonts, and emoji themes are available through the project's web-based asset generator.
Technical Details
Built with ESP-IDF on ESP32-S3. Uses the OPUS audio codec, WebSocket or MQTT+UDP for cloud communication, and integrates with the xiaozhi-esp32-server backend for device management.
