exo

Link everyday devices into one cluster and run models too big for any single machine.

Install Repository Compare

Fit.

Catalog snapshot

Platforms: Linux, Macos
Operating systems: Linux, macOS
Ram floor: 16 GB
Gpu: Optional · 8 GB VRAM helps
Docker: No
Setup: Intermediate
Deploy as: Local
Commercial use: Allowed

Clusters of 8–16 GB machines beat one box for 30B+ models.

Project.

Fetched Sep 12, 2026¹

License: Apache 2.0
Stars: 47.4k¹
Forks: 3.5k¹
Open issues: 200¹
Language: Python¹
Last commit: Aug 25, 2026¹
Tags: clusterggufapple-silicon

About.

exo networks phones, Macs, and Linux boxes into one inference cluster and splits a model across them. It is the cheapest way to run models larger than any single machine you own. Expect tinkering, not a product.

device clustering · ChatGPT-compatible API · Apple Silicon · model sharding

Alternatives.

Nearby.

All Local AI

LllamafileSingle-file LLM runtime from Mozilla. Download, chmod, run.Compare
GgptmeMIT terminal agent toolkit with code, shell, and web tools — local-model friendly.Compare
GGPT4AllNomic’s local chat app and model catalog for laptops without a GPU.Compare
Ttext-generation-webuiGradio UI for local models with loaders, LoRA, and extensions.Compare
LLiteLLMProxy that translates many model APIs into one OpenAI-shaped client.Compare