Cactus
Source-available inference engine for running models on phones, wearables, smart-home and robotics hardware, shipping its own quantization, kernels and runtime behind OpenAI-compatible text, speech and vision APIs.
What it does
- Model inference
Features 6
- Cactus EngineC inference API for chat completion, streaming, tool calling, transcription, embeddings, RAG, vision and cloud handoff.
- Cactus GraphC++ computation graph for tensor operations, matrix multiplication, attention, normalization and activations.
- Cactus KernelsARM NEON SIMD kernels for matmul, attention, convolution, quantization, DSP and image processing.
- Cactus QuantsRotation-and-codebook quantization from 4-bit down to 1-bit for all weight tensors.
- Cactus HybridRoutes hard queries to the cloud automatically based on the local model's confidence.
- Python package and CLIPython package and command-line tool for running, downloading and converting models.
Sources
- Pricing
-
Official documentation · 18 Sep 2026
Licence text read in full; the two $2,000,000 thresholds and the thirty-day commercial-licence clause are quoted from it.
- Description
-
Official documentation · 18 Sep 2026
Documentation's own top-level sections: Cactus Engine ('OpenAI-compatible APIs for text, speech, and vision'), Cactus Graph, Cactus Kernels, Cactus Quants.
- Sold within
-
Official documentation · 18 Sep 2026
Obtained and run directly from the repository under its own licence; no parent offering is required.
- URL
-
Official documentation · 18 Sep 2026
Repository home, fetched 2026-09-18: 'Quantization, kernels, runtime and inference engine for mobiles, wearables, smart home and robots.' 6,020 stars, not archived, last pushed 2026-09-08.
- Interfaces
-
Official documentation · 12 Oct 2026
Page: 'Python package and CLI / OpenAI-compatible APIs for text, speech, and vision'.
- Modalities
-
Platform listing · 14 Oct 2026
PyPI listing published by Cactus Compute: 'On-device AI inference — LLM, vision, speech, embeddings, and RAG'.
- Intended for
-
Official documentation · 13 Oct 2026
Page: 'Python package and CLI'.
Also from Cactus Compute
-
Cactus Platform Platform
Hosted fine-tuning and evaluation service for Cactus models, running Cactus datasets, 2-bit quantisation, evaluation tracking and full-depth fine-tunes on Cactus Compute's own training infrastructure.
Open Cactus Compute in the directory
Something wrong here? Send a correction — quote this product id: cactus.