ElatoAI Turns a $45 Chip Into a Real-Time Voice AI Device
ElatoAI is an open-source platform that turns a $5 ESP32-S3 chip into a real-time voice AI device supporting 100+ models from OpenAI, Gemini, and more
Takeaways
- Full-stack open-source platform turns an ESP32-S3 into a real-time voice AI device with 100+ model support, OTA updates, and a web dashboard.
- Supports major voice AI APIs including OpenAI Realtime, Gemini Live, xAI Grok Voice, ElevenLabs, and Hume AI EVI-4 out of the box.
- Sub-2-second global latency achieved via Deno/Cloudflare edge functions and Opus audio codec at 12kbps/24kHz.
- Hardware costs are minimal: ESP32-S3 + I2S mic + amplifier; pre-assembled dev kit available for $45 at elatoai.com.
- Key limitations: 2-3s cold start, ~20-minute session cap, no mid-sentence interruption on device, and ongoing API costs (~$0.20/15 min with OpenAI).
- Featured in OpenAI Cookbook and covered by Wired, Adafruit, and Hackster; nearly 1,900 GitHub stars and active community on GitHub.
ElatoAI is a fully open-source platform for building real-time voice AI devices on an ESP32-S3 microcontroller, the kind of cheap, thumb-sized chip common in hobbyist electronics kits. The project ships everything: ESP32 firmware in C++/Arduino, a Next.js web dashboard, Deno edge functions, OTA firmware updates, and device management. You bring the hardware and API keys.
It supports instant speech-to-speech conversion through OpenAI's Realtime API, Gemini's Live API, xAI's Grok Voice Agent API, ElevenLabs Conversational AI Agents, and Hume AI EVI-4. Beyond those native pipelines, you can run 100+ STT, LLM, and TTS combinations through a FastAPI server with Pipecat, or deploy through Cloudflare Workers for access to 80+ LLMs, 10+ TTS models, and 5 STT models.
The gap it fills
The author built ElatoAI after failing to find a reliable, secure WebSocket speech-to-speech service with a server-side component for business logic. OpenAI's embedded repo sets up WebRTC with ESP-IDF, but it's not beginner-friendly and stops at the device layer. ElatoAI uses the Arduino framework instead of ESP-IDF, which is far more accessible, and routes audio through edge functions that handle authentication, device management, and LLM API calls. The project has been featured in the OpenAI Cookbook, Adafruit, Hackster, and Wired, and has nearly 1,900 GitHub stars.
How it's built
The architecture has three layers:
- ESP32-S3 firmware (C++/Arduino): Captures audio via an I2S microphone (e.g. INMP441), encodes it with the Opus codec, and streams it over a secure WebSocket to the edge server. No PSRAM required.