78/xiaozhi-esp32

XiaoZhi is a DIY AI chatbot that runs on ESP32 microcontrollers, tiny, inexpensive chips (often costing just a few dollars) used for building physical hardware projects. The result is a voice-enabled AI assistant you can build yourself using either a breadboard and components, or one of 70+ supported off-the-shelf development boards. The chatbot connects to large AI models like Qwen or DeepSeek over Wi-Fi, enabling natural voice conversations with real AI intelligence in a pocket-sized physical device. It supports offline wake-word detection (so it listens for its name without sending audio to the cloud), has an optional OLED/LCD display that shows emoji reactions, and can recognize individual speakers' voices. What makes it especially interesting is MCP (Model Context Protocol) integration, a standardized way for AI models to control external devices and services. This means the chatbot can control smart home devices, interact with your computer desktop, search for information, or manage hardware peripherals like LEDs and motors, all through voice commands. For a hardware-curious vibe coder or maker, this is the kind of project you can assemble in an afternoon with a development board like the M5Stack CoreS3 or ESP32-S3-BOX3. Personal users can register a free account on xiaozhi.me and use it with the Qwen real-time AI model at no cost. Built in C++ for the embedded firmware, with documentation available in English, Chinese, and Japanese. It's an active, well-supported project that bridges physical maker culture with modern conversational AI.