Overview
dots runs one large language model across several Apple Silicon Macs, so a team can use a model none of its machines could load alone. It splits the model layer by layer across the pool and serves the whole pool as one OpenAI-compatible endpoint, so any app that talks to ChatGPT-style services can use it.
I built dots with my friend Arpan. He built the inference engine. I designed the product and built the app on top of it: the view of the pool, the chat, and lending a Mac to the pool from an invite link. Today it runs our own Macs as one pool, and we have pitched it to investors.