Cactus Compute
1 Tool · 2 Intel
Cactus Compute, Inc. is a Y Combinator-backed AI infrastructure company building quantization, kernels, and inference engines that let language, vision, and speech models run locally on phones, wearables, and other edge devices. Led by Henry Ndubuaku, the company maintains its work as open source under the cactus-compute organization on GitHub and Hugging Face. Its compact Needle models are built for on-device tool-calling and structured extraction rather than cloud inference.
Is this you? Sign in with X to claim this profile.
Tools
14MB on-device foundation model for tool calling and structured extraction, ~28MB RAM per session.
Intel
Cactus's Needle 3 ships 25-121M parameter, 2-bit tool-calling models as 8-29MB binaries, using a Monarch Hadamard MLP to cut FFN cost, and beats LFM2.5, Qwen3.5, and Apple's on-device model on a mobile-action benchmark.
Cactus releases Whistle, an open speech-to-text model in a single 16.9 MB file. It transcribes seven languages with word timestamps and embeddings, runs CPU-only, and shares its C++ engine with Needle for on-device speech-to-tool-call pipelines.