Cactus Compute - Sub-50mb Super Intelligence For Consumer Devices

Cactus runtime

Prebuilt runtime binaries for Needle 3, Whistle and Mirage. One engine runs text, speech and vision locally. Model weights live in their own repositories.

Get started

pip install -U cactus-compute

Use import cactus in Python. The package downloads the matching engine and model weights on first use. See the Python API and source code.

To download a native runner with a model, choose a platform from the table below.

cactus download macos-arm64 --model whistle --out ./runtime
./runtime/macos-arm64/cactus --model ./runtime/macos-arm64/whistle.cact --audio clip.wav

Use --model needle3 or --model mirage for the other models. Omit --model to download binaries alone.

Supported platforms

Release 4.0.0 has 17 target folders under releases/4.0.0. Native runners load .cact weights at runtime. Static libraries include cactus.h for C and C++ applications.

Target Folder Files
macOS Apple Silicon macos-arm64 cactus, libcactus.a, cactus.h
Linux x86-64 linux-x86_64 cactus, libcactus.a, cactus.h
Linux ARM64 linux-arm64 cactus, libcactus.a, cactus.h
Linux ARMv7 linux-armv7 cactus, libcactus.a, cactus.h
Linux RISC-V linux-riscv64 cactus, libcactus.a, cactus.h
Linux MIPS32el linux-mipsel cactus, libcactus.a, cactus.h
Windows x64 windows-x86_64 cactus.exe, libcactus.a, cactus.h
Windows ARM64 windows-arm64 cactus.exe, libcactus.a, cactus.h
Android ARM64 android-arm64 cactus, libcactus.a, cactus.h
Android ARMv7 android-armv7 cactus, libcactus.a, cactus.h
Android RISC-V android-riscv64 cactus, libcactus.a, cactus.h
iOS ARM64 ios-arm64 libcactus.a, cactus.h
iOS Simulator ARM64 ios-sim-arm64 libcactus.a, cactus.h
tvOS ARM64 tvos-arm64 libcactus.a, cactus.h
watchOS ARM64 watchos-arm64 libcactus.a, cactus.h
Browser and Node.js wasm cactus.js, cactus.wasm, cactus.h
WASI Preview 2 wasm-component cactus.component.wasm, cactus.wit

The Python runtime wheels cover macOS ARM64 and x86-64, Linux ARM64 and x86-64 with glibc or musl, and Windows ARM64 and x64. These contain the shared library. Install cactus-compute from PyPI to use them through the Python API.

The WASI component exports cactus:runtime/engine@4.0.0. It is also available as the signed OCI artifact ghcr.io/cactus-compute/cactus-runtime-component:4.0.0.

Weights and releases

Release artifacts have SHA-256 checksums in the manifest. For reproducible downloads, use revision 1a028f5a3972b67dd8d8752f882113d475e7ec4c or tag v4.0.0 and paths under releases/4.0.0/.

See the deployment guide for native integration, browser use and offline setup.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Collection including Cactus-Compute/cactus