Token budget analyzer for AI context windows — like du for tokens
The context compression protocol for LLM inference. Eliminate 93% token redundancy in one line.