Mind Lab

Maker of the Macaron V1 Mixture-of-LoRA agent models.

Mind Lab builds the Macaron family, agent models designed around continual learning rather than a single frozen set of weights. Each Macaron V1 release pairs a base model with four LoRA specialists covering chat, personal-agent work, coding and generative UI, with a router picking the specialist per request. The line ships in two sizes, the 748 billion parameter Venti and the 50 billion parameter Tall, the latter built to run locally.