Cactus Compute SDK (Python) — cactus_init

Initialize local AI models for low-latency inference on mobile and desktop using the Cactus Compute Python SDK.

The cactus_init function is the core entry point for loading quantized AI models into the Cactus Engine for local execution. It maps weights directly from disk using zero-copy memory mapping, supporting GGUF and native Cactus formats for models like Qwen3 and Llama. By specifying the model path and an optional RAG corpus, you can achieve sub-120ms latency on ARM and Apple Silicon hardware. This Python interface provides a high-performance bridge to the C++ backend, enabling private, offline inference for text, vision, and speech tasks.

https://cactuscompute.com