/dev/world 2026

Apple Platform Architect

Local AI inference on Apple Silicon is fast. Two Mac Studios running together are faster - and this session shows you exactly how to make that happen. This talk walks through building a real, working distributed inference cluster using Apple's MLX framework, then connecting it to a Swift agentic app with full tool-calling support. Attendees will see a live two-node cluster (ring-backend over Ethernet, no Thunderbolt required) serving a sharded large language model via mlx_lm.server's OpenAI-compatible endpoint - and a Swift client hitting that endpoint with the same tool-calling patterns they'd use against Apple's on-device SystemLanguageModel.
 The central lesson: once your cluster is running, your Swift app doesn't need to change. The distribution lives in the server layer; your application code stays clean, portable, and ready to swap backends. Attendees will leave knowing how to configure a multi-node MLX hostfile, launch distributed inference with mlx.launch, and wire up tool-calling agents in Swift against a locally hosted cluster — no cloud account, no API costs, no data leaving the room.

About the presenter

Ace Authors is a Melbourne-based iOS/macOS developer with 10+ years building software for NAB, ANZ, ME Bank, Officeworks, and Coles Myer. He holds a Computer Science degree from QUT and a Master's from the Asian Institute of Technology. As Apple Platform Architect at RMIT University, Ace explores emerging Apple platform technologies and helps architect research initiatives.