78/xiaozhi-esp32

This is an open-source firmware project that turns ESP32 microcontrollers into conversational AI endpoints using the MCP protocol. In practice, that means you can flash a cheap, widely available chip and have a working voice or text chat interface running locally, without relying on a cloud-hosted device layer. The project handles the low-level integration work — connecting the microcontroller to a language model backend, managing audio input and output, and maintaining conversation state — so builders can skip the most tedious embedded plumbing and focus on the application layer. The appeal is the hardware cost. ESP32 modules run a few dollars each, making this relevant for anyone prototyping smart home devices, kiosk interfaces, industrial terminals, or physical AI companions without a large hardware budget. The MCP foundation also means it can slot into broader tool-calling workflows. The honest reservation is accessibility. Documentation is primarily in Chinese, the community is concentrated in that ecosystem, and debugging embedded firmware without strong documentation support is genuinely painful for outsiders. -> Best for: Hardware-leaning indie hackers building physical AI products on a tight budget.