ShrinkML
Builds a small AI model that runs on Apple devices.
What it is for
ShrinkML turns a task you describe in plain language, together with your own data, into a small model that runs on Apple devices, and the Swift code that uses it. It is for apps that need one narrow job done well on the device itself.
Who it is for
Teams that want AI on the device in an Apple app, without running a machine-learning operation of their own.
How it works
You describe the task, and ShrinkML turns it into a structured list of the fields to find.
It trains the model on your data, in the cloud or on your own hardware, and tests it on examples the model never trained on.
You get the model and the Swift code that runs it, with accuracy measured again at each precision you compile to.
Facts
- Training
- In the cloud, or on your own Mac with Apple silicon
- Runs on
- Apple devices, on the device itself
- You get
- The model, and the Swift code that runs it
- Availability
- Closed testing, one company at a time. Public beta planned for Q4 2026.