Company Page

TurboFieldfare is an open-source, purpose-built Swift and Metal inference engine that runs Google DeepMind's Gemma 4 26B-A4B model in approximately 2 GB of RAM on any Apple Silicon Mac. Rather than relying on compression or aggressive quantization, it keeps only the shared model core in RAM and streams routed expert weights directly from SSD during inference. The project includes a native macOS app, a command-line interface, and an experimental OpenAI-compatible local server, making powerful on-device AI accessible even on entry-level 8 GB MacBook Air devices.

Company Founded in 2026

SaaS