Skip to content

exo

Run AI models on the devices already at home

Visit official site

What exo is

The project is described as experimental software that unifies existing home devices such as iPhone, iPad, Android, Mac and Linux into one GPU for running AI models. It supports models like LLaMA and includes dynamic model splitting plus a ChatGPT-compatible API.

What you can do with it

  • Run popular models such as LLaMA across your devices
  • Split a model dynamically by network topology and device resources
  • Discover devices automatically with no manual configuration
  • Use multiple inference engines including MLX, tinygrad and llama.cpp

Application scenarios

  • An iPhone and an iPad form a cluster for image recognition
  • Several Linux devices at home run natural language processing models
  • A Mac and an Android device jointly handle complex data analysis

Getting started

  • Install from source by entering the exo directory and running pip for dependencies
  • Run exo on multiple devices following the documentation examples
  • Integrate the provided API into an application to run models
  • Open port 8000 on a device for the ChatGPT-style web interface

Pros & cons

✓ What we like

  • Multiple inference engines are supported, including MLX, tinygrad and llama.cpp
  • Network modules cover GRPC, Radio and Bluetooth

! What to watch out for

  • The project is described as experimental
  • Model splitting depends on the available network topology and device resources

FAQ

Which models are named?

The material names LLaMA among the popular models supported.

How do apps connect to it?

The feature list states a ChatGPT-compatible API needs only a one-line change in an application.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review