Cactus

Cactus is an on-device AI engine for developers adding automation to mobile apps, wearables and other small devices. Its Needle model turns requests into tool calls locally, including when the device has no internet connection.

• Local automation: Needle handles tool calling with compact 2-bit models designed for devices with limited memory, including smart home hardware and microcontrollers.
• Chat, vision and speech: The engine supports chat with streaming responses, image input and speech transcription, with tested model families including Gemma, Qwen, Liquid, Whisper and Parakeet.
• Search over local data: Embeddings, a vector index and retrieval-augmented generation let applications use local information when answering requests.
• Optional cloud handoff: Hybrid routing can send difficult queries to a cloud service based on the local model's confidence; requests routed this way leave the device and need connectivity.

Cactus runs on iOS, Android and Mac. Bindings include Swift, Kotlin, Flutter and React Native, plus Python and Rust for other application integrations. The engine uses ARM-optimized kernels and model quantization to reduce inference demands on small hardware. Local inference runs on the user's device; cloud handoff is a separate, optional path.