exo-distributed Skill

"Run models across heterogeneous devices by forming GPU clusters with zero configuration."

Trit: 0 (ERGODIC - coordination/orchestration) Color: Neutral (60-180° hues) Source: Random walk fusion over DuckLake interactions + DeepWiki exo-explore/exo

Overview

exo enables distributed LLM inference across multiple Apple Silicon devices:

• Auto Peer Discovery: Devices find each other automatically
• RDMA over Thunderbolt 5: Low-latency direct memory access
• MLX Backend: Native Apple Silicon acceleration via mlx.distributed
• Pipeline + Tensor Parallelism: Shard models across devices

Quick Start