cactus
Cactus is a Python package that runs tool-calling, speech and image models locally on your own device for developers building private, small-footprint agents.
What it is good for
- Running AI models on a device without sending data to the cloud
- Transcribing speech from audio files without an online service
- Describing images and extracting structured data from photos
- Letting a local model call your own functions as tools
At a glance
- Who it is for
- Setup difficulty
- Developer ยท Requires programming
Level 5 of 5. You need to write code to use it.
- Cost
- Free and open source
- Type
- Framework or SDK
- Tasks
- Works with
- Python, MLX, macOS
How to get started with cactus
- Install Python 3.9 or later.
- Run
pip install -U cactus-computein a terminal. - Write a Python function and mark it with
@cactus.toolso the model can call it. - Run your script; the engine and model weights download on first use.
From the project's own documentation. If a step does not match what you see, the official pages have the latest instructions.
Before you try it
- You need to write Python code to use it.
- Model weights and the engine download and cache on first use.
- The Mac MLX feature needs Apple silicon, macOS 26 or newer, and Python 3.12 or newer.
- Anonymous Python usage counts are collected unless you set `CACTUS_TELEMETRY=0`.
Momentum
- Rank this week
- #21 in builder tools
- Best rank
- #21 week 46, 2026
- Stars gained last week
- +158
- GitHub stars
- 14,764
GitHub numbers as of Nov 16, 2026. Updated weekly.
Ranking history
Official links
Questions
- Is cactus free?
- Yes. It is free and open source (Apache-2.0 license).
- Do I need to know how to code to use cactus?
- Yes. You need to write code to use it.
- What does cactus work with?
- According to its own documentation, it works with Python, MLX, and macOS.
- How do I get started with cactus?
- 1. Install Python 3.9 or later. 2. Run pip install -U cactus-compute in a terminal. 3. Write a Python function and mark it with @cactus.tool so the model can call it. 4. Run your script; the engine and model weights download on first use.
- What should I know before trying cactus?
- You need to write Python code to use it. Model weights and the engine download and cache on first use. The Mac MLX feature needs Apple silicon, macOS 26 or newer, and Python 3.12 or newer. Anonymous Python usage counts are collected unless you set `CACTUS_TELEMETRY=0`.
- What is cactus good for?
- Running AI models on a device without sending data to the cloud, transcribing speech from audio files without an online service, describing images and extracting structured data from photos, and letting a local model call your own functions as tools.
Similar agents to cactus
- openai-agents-python
Builds multi-agent workflows in Python, while this focuses on running small models locally on your device.
- litellm
Routes calls to many hosted model providers through one interface instead of running models locally.
- LangChain
A broader framework for connecting models and data sources, not centered on small on-device models.
- crewAI
Focuses on teams of agents splitting up work, rather than local inference for speech, vision and tool calls.
- headroom
Shrinks tool outputs to cut token costs for coding agents rather than running models itself.
- cognee
cognee is a memory platform that stores documents, code, and conversations as a searchable knowledge graph so AI agents remember context across sessions, for developers.