What OpenToys Provides
OpenToys is the local-first platform version of ElatoAI for toys, devices, and robots. It launched on 2026-03-14 as a fully open-source project that requires no cloud connection or recurring fees.
All processing stays on the device. Voice data never leaves the hardware.
Supported Hardware and Software Stack
The platform targets ESP32-S3 microcontrollers connected over websocket for audio transport. It also runs on Apple Silicon Macs.
The software stack combines Rust and React for the application layer with Whisper for speech recognition, Qwen3-TTS for speech synthesis, and MLX models for the language component.
- ESP32-S3 hardware for embedded deployment
- Whisper ASR for on-device transcription
- Qwen3-TTS and MLX LLMs for local generation
Key Features for Custom Builds
OpenToys supports more than ten languages including English, Chinese, Spanish, French, Japanese, Korean, Portuguese, German, and Italian. Voice cloning works from less than ten seconds of reference audio.
Developers can customize the entire pipeline to create companions, interactive toys, or simple robots. The project remains community-driven with source code available for inspection and modification.
Privacy and Safety Considerations
Because inference runs locally, conversations stay on the device. No data is sent to external services, removing the usual privacy risks associated with cloud voice assistants.
Users should note that LLMs and TTS systems can still hallucinate or generate inappropriate responses. The project documentation stresses that AI does not replace human interaction, particularly around children, and requires ongoing parental oversight.