tl;dr: TurboFieldfare is a Swift/Metal runtime that runs Gemma 4 26B-A4B (a MoE model with ~3.88B active params) in about 2 GB of RAM on Apple Silicon Macs, including 8 GB models, by keeping shared weights and KV cache resident while streaming individual experts from SSD per token. Measured decode speeds range from 5-6 tok/s on an M2 Air to 31-35 tok/s on an M5 Pro. It ships with a native Mac app, CLI, and a loopback OpenAI-compatible server, and requires macOS 26 with Metal 4.