An Identity for Forward-Deployed Reinforcement Learning

Osmosis is forward-deployed reinforcement learning: it post-trains task-specific models that beat the frontier foundation models at a fraction of the cost, across document extraction, multi-tool agents, and domain code generation. The category it lives in (post-training, GRPO, DAPO) reads as jargon to the teams who need it most, and the name's own idea, intelligence absorbed gradually through contact, wasn't yet carrying any weight. The identity had to make a deeply technical practice feel inevitable rather than arcane.

An identity built on the osmotic principle: intelligence that diffuses across a membrane and stays. The visual language draws from Huang Gongwang's Dwelling in the Fuchun Mountains, the 14th-century ink scroll where mist, water, and stone resolve into one porous, unbounded system. That brush-and-wash lineage carries the argument: as mountains and rivers resolve through gradients of ink, meaning resolves through flows of data, signals, and context. The result is a restrained technical wordmark and type system that holds from a research note to an enterprise deck. Gradient-as-argument, where absorption is the whole story, intelligence as permeable as mist, seeping into a company's own data and compounding there.