exo (Exo Labs)

exo is an open-source project from Exo Labs that lets you run frontier-scale AI models entirely locally by stitching together the memory and compute of every Mac, workstation, or device on your network into one distributed inference cluster. Where a single laptop can comfortably run an 8B or 14B model, exo's peer-to-peer runtime can spread a Llama-3 70B, DeepSeek-V3, or Qwen 32B model across several Apple Silicon Macs connected by Thunderbolt or local network, and serve it through APIs that drop in for OpenAI, Anthropic, and HuggingFace clients.