ElatoAI provides realtime voice AI for ESP32 devices using 100+ models via Deno Edge or Cloudflare Workers, with secure WebSockets and OTA updates. It integrates multiple AI providers and an ESP32 hardware stack, aiming for low-latency global conversations.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Realtime Voice AI models running on Arduino ESP32 with Secure WebSockets and edge functions, enabling voice-powered devices and toys with 100+ models across OpenAI, Gemini, xAI Grok, ElevenLabs, Hume, and more. It includes a frontend Next.js app, edge server components, and an ESP32 IoT client.
How it works
The stack comprises three components: a Frontend Client (Next.js, hosted on Vercel), Edge Server Functions (Deno Edge or Cloudflare Workers), and the ESP32 IoT Client (PlatformIO/Arduino). ESP32 devices connect to the edge server via Secure WebSockets, streaming audio in Opus format. The edge functions route to various LLM/TTS/STT providers, enabling real-time speech-to-speech pipelines. Assets include a mermaid-based architecture and a project structure showing frontend, server, and ESP32 components communicating with a Supabase DB.
Getting started
The README provides quickstart sections and documentation links:
- Quick Start: https://www.elatoai.com/docs/quickstart
- Build with PlatformIO: https://www.elatoai.com/docs/platformio
- Build on Arduino IDE: https://www.elatoai.com/docs/arduino
- Deploy globally: https://www.elatoai.com/docs/blog/deploying-globally
- Deploy multiple devices: https://www.elatoai.com/docs/blog/multiple-devices
Exact installation or setup commands are not included in the provided excerpt beyond these links.
Recent releases
The latest releases section states:
- latest 0: none
No versioned release notes are listed in the provided data.
Traction
The repository has 1871 stars and 233 forks. Open issues: 6. Language: TypeScript. License: none listed. Created: 2025-04-08. Last push: 2026-07-10.
Since stars_7d or stars_1d aren’t provided in the data, this section is omitted per the facts-only rule.
Behind the repo
No startup or company link is provided beyond the project itself, so this section is omitted.
Caveats
License is listed as none. Other constraints mentioned include:
- 2-3s cold start time while connecting to edge server
- Tested with up to 17 minutes of uninterrupted conversations
- Edge server stops when wall clock time is exceeded
- No speech interruption detection on ESP32
Public security notes include Secure WebSockets, optional API Key encryption with 256-bit AES, and Supabase DB for authentication with Postgres RLS.