llmtrim 0.9.2

Static, deterministic LLM prompt/payload compressor — cut input tokens 30-90% with zero extra model calls
Documentation