Hardware and Software Stack
OpenToys runs inference on Apple Silicon Macs using Rust and React while targeting ESP32-S3 hardware for the toy or robot side.
The stack includes Whisper for automatic speech recognition, Qwen3-TTS for speech synthesis, and MLX-based LLMs for on-device reasoning.
- Full support for English, Chinese, Spanish, French, Japanese, Korean, Portuguese, German, Italian and additional languages
- Voice cloning from less than ten seconds of audio
The bottom line: OpenToys runs inference locally with Whisper ASR, Qwen3-TTS, and MLX LLMs while supporting voice cloning and multilingual output.
Firmware Flashing Steps
Connect the ESP32-S3 board to the Mac.
Open the OpenToys Settings interface and select the target device to flash the firmware image.
Network and Runtime Configuration
After flashing, the ESP32-S3 creates a WiFi captive portal named ELATO.
Connect to the portal to enter local network credentials so the device can maintain its connection when powered on.
Safety and Usage Limits
LLMs and TTS models can generate hallucinations or inappropriate responses.
OpenToys documentation states the system should not replace human interaction.