Local AI Overview

OpenToys is the local-first version of Elato. Your Apple Silicon Mac runs every model, and your ESP32-S3 streams audio to it over a websocket. Nothing leaves your home.

OpenToys on GitHub

Download the app, flash the firmware and start building.

The local stack

Stage	Model	Page
STT (speech to text)	Whisper large-v3-turbo	STT
LLM (the brain)	Any `mlx-community` model, default Ministral-3-3B (Gemma 4, Qwen3, Llama, gpt-oss and Muse Glimmer also supported)	LLM
TTS (text to speech)	Qwen3-TTS by default, plus nine other backends	TTS
Pipeline	STT → LLM → TTS over a websocket	Realtime

All models run through MLX, Apple’s machine learning framework, so you need an Apple Silicon Mac (M1 or newer).

Cloud or local?

	Cloud AI	Local AI
Cost	Pay per API usage	Free
Privacy	Audio goes to the provider	Stays on your Mac
Latency	Native realtime models	Cascaded STT → LLM → TTS
Hardware	ESP32 + internet	ESP32 + Apple Silicon Mac

Looking for OpenAI, Gemini, Grok, Eleven Labs or Hume? See Cloud AI.

Get started

Install OpenToys

Download the latest DMG from GitHub Releases.

Flash your ESP32-S3

In OpenToys, open Settings and pick Connect your ESP32 Here, then click Flash. Wiring is covered in the firmware guide.

Join the toy's WiFi and press Play

The toy hosts an open network called ELATO. Join it from your Mac and press Play.

While your Mac is on ELATO it has no internet. Download models and voices before joining.

OpenToys on GitHub

​The local stack

​Cloud or local?

​Get started

The local stack

Cloud or local?

Get started