Thanatos9404

Context engine that compresses older tool output and turns locally with LLMSlim instead of asking a model for a summary: no LLM call, deterministic output, and the start of the prompt is sent unchanged, so prompt caching keeps working. Select it with context.engine: llmslim.