Exo

Exo is an open-source project with 47,900+ stars on GitHub that runs frontier AI models at home by clustering the devices you already own — Macs, PCs and phones — into one personal AI cluster.

Overview

A single laptop rarely has enough memory for the largest open models. Exo’s answer is to stop thinking in single machines: install it on several devices and it discovers them automatically, splits each model across the cluster with topology-aware partitioning, and serves the combined system through a ChatGPT-compatible API. On Apple silicon it uses the MLX engine; elsewhere it builds on llama.cpp, so a model too big for any one device can run comfortably across two or three. The cluster rebalances as machines join and leave, and popular open models from Hugging Face are supported on day one. Everything stays on your home network — no cloud account, no rental GPUs, no data leaving the house. Exo is Apache-licensed and actively developed by the exo team and community.

Key Features

• Cluster your own hardware – combine Macs, PCs and other devices into a single inference cluster.
• Automatic device discovery – nodes find each other with zero configuration.
• Topology-aware sharding – models are split across devices according to memory and connection speed.
• ChatGPT-compatible API – point existing chat apps and tools at your home cluster.
• MLX and llama.cpp engines – fast paths for Apple silicon and general hardware alike.
• Day-one model support – run newly released open models from Hugging Face immediately.
• Fully private – inference happens at home; nothing is sent to a cloud provider.

💳 Pricing

Exo is free and open source under the Apache License 2.0. Your only cost is the hardware you already own and the electricity to run it.

Who Is It For

AI hobbyists, researchers and privacy enthusiasts who want to run large open models locally and would rather pool the devices on their desk than rent cloud GPUs.