# cactus-compute/cactus

Quantization, kernels, runtime and inference engine for mobiles, wearables, smart home and robots.

Repository: https://github.com/cactus-compute/cactus
Canonical: https://ross.abutalabs.com/products/cactus
Homepage: https://cactuscompute.com
Language: C++
License: NOASSERTION
License Family: other
Topics: android, framework, ios, llamacpp, llm, llm-inference, llms, transformer, ai, edge, mobile, smartphone, speech, whisper, arm, edge-ai, mobile-inference, on-device-ai, quantiz, rag
Last push: 2026-08-26T16:45:30+00:00

## Health v2 (maintenance only)
Score: 86/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 98, longevity 35
- inputs: {"age_days": 497, "days_push": 7, "days_rel": 16, "gap_med": 7, "n_releases_24m": 20}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 5934, forks 495 (observed 2026-08-28T04:09:32.522675+00:00)

## What it is
Cactus is a hybrid edge-cloud AI inference engine for mobile devices, wearables, smart home devices, and robots, built in C++ with custom quantization, CPU/GPU kernels, and a zero-copy computation graph. It exposes OpenAI-compatible APIs for text, speech, and vision, with automatic cloud handoff when on-device models are uncertain.

## Use cases
- run LLM inference on-device on Android and iOS
- deploy AI models on wearables and microcontrollers
- on-device speech transcription with Whisper
- build RAG apps that run locally on phones
- tool calling and structured extraction on tiny edge devices
- quantize transformer models for mobile inference

## When to choose
- you need low-latency, battery-efficient LLM inference on smartphones or embedded hardware
- you want on-device AI with automatic cloud fallback for uncertain predictions
- you need a single engine covering text, speech, and vision at the edge

## When to avoid
- you need a fully permissively licensed dependency (license is non-standard)
- you only target server or desktop GPUs with no edge constraints
- you need a mature ecosystem with broad model support like llama.cpp or ONNX Runtime

## Facets
- artifact type: framework
- maturity: active
- function: llm-inference, rag, speech-recognition, machine-learning, sdk
- domain: machine-learning, large-language-models, artificial-intelligence, mobile-development, embedded-systems, speech-processing
- platform: cross-platform, cpp, embedded
- tags: on-device-ai, edge-ai, quantization, wearables, llamacpp, whisper, cloud-fallback, inference-engine, arm, android, ios, macos, mobile

## Member repositories
- cactus-compute/cactus (main) score 86

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:32.522675+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:50:47.239415+00:00, confidence not recorded.
  - readme: https://github.com/cactus-compute/cactus (fetched 2026-08-28T04:09:32.522675+00:00, sha 092a9982ad14)
  - homepage: https://cactuscompute.com (fetched 2026-08-29T08:46:24.059457+00:00, sha e3053bc852ab)
  - site_page: https://docs.cactuscompute.com (fetched 2026-08-29T08:46:24.063108+00:00, sha 36c6c3c2e4f9)
- Data as of 2026-08-30T08:39:29.467469+00:00.