cactus

Open-source engine by Cactus Compute that runs small AI models for text, speech and images locally on consumer devices.

Website ↗

cactus is an inference engine by Cactus Compute for local AI on consumer devices, under the Apache License 2.0. It runs the company's own models: Needle for tool calling, structured extraction and embeddings, Whistle for speech transcription in seven languages, and Mirage for images. The documented way to use it is the Python package cactus-compute with its command-line tools; on a Mac it can also run language models through MLX.1

In the browser

There is also a WebAssembly build: cactus.js and cactus.wasm. The JavaScript file creates the module, the model weights are copied into its memory and loaded with cactus_load, and cactus_describe takes the RGB pixels of an image and returns JSON. The build I took from a demo needed a missing semicolon in cactus.js before it ran.2


  1. cactus on GitHub. ↩

  2. Project notes, 18 November 2026. ↩

More on this

Used in
  • What the Machine Sees A small app from my old website: choose a photo, and the vision model Mirage found objects, text and the scene in it, entirely in the browser.
By
  • Cactus Compute Company behind cactus, an engine for running small AI models on consumer devices, and models such as the vision model Mirage.
See also
  • Mirage Small vision model by Cactus Compute that finds objects with outlines, text and the scene in a photo and returns them as JSON.
Tags

Show in the graph view →