ExecuTorch is PyTorch's open-source toolkit for running AI models on the device itself, such as a phone, headset, laptop, browser or tiny chip, instead of on a server.
In plain words
You start with a model built in PyTorch and capture it with torch.export. ExecuTorch then prepares a separate .pte file for each kind of hardware you target, and a small runtime runs that file. Parts of the model can be handed to a fast chip such as a GPU or NPU, with ordinary processor code as a fallback. Apps call it from C++, Python, Swift, Kotlin or JavaScript. Meta uses it for on-device features in Instagram, WhatsApp, Facebook and Messenger.
A quick example
Imagine a note-taking app that sums up your voice memos. With ExecuTorch, the summary model ships inside the app and runs on the phone, so the recordings never have to leave it.
Where you'll meet it
- Phone and wearable apps that run AI without a server.
- Language models on laptops and in web browsers.
- Small embedded devices and microcontrollers.
Related
Sources
- Docspytorch/executorch, PyTorch (GitHub) · read 10 Nov 2026
- DocsExecuTorch documentation, PyTorch · read 10 Nov 2026
- DocsExecuTorch WebGPU: PyTorch Models on the GPU, in the Browser, PyTorch blog · read 10 Nov 2026
Checked by WhatIsAI against the sources above · 10 Nov 2026. Spotted an error? Report it.