ElatoAI

Why we included this project

Most voice agents assume a beefy cloud server is sitting behind them, but this project runs the whole realtime speech loop on an ESP32. It chains speech-to-text, an LLM, and text-to-speech over secure websockets, and the edge functions plus a Pipecat/FastAPI backend keep sessions alive well past twenty minutes rather than dropping the connection mid-conversation. Anyone building a talking gadget, whether it is an AI toy, a companion, or some other interactive device, gets working recipes for hooking the microcontroller up to dozens of cloud voice models, so you are not hand-writing every integration yourself. OpenAI's cookbook references it as an example of running their Realtime API on Arduino, which is a practical endorsement from the people who make that API. If you are prototyping a device that should hold an actual spoken conversation, studying this project beats building your own stack from scratch.