MinML
MinML: A local, universal token cache with safe prompt compression, cut LLM costs 30–60% without losing meaning, across any model or tool doubling context and performance.
Core trick
Compresses prompts against a local token cache so repeated context is not resent to the model.