actrone-memory
    Preparing search index...

    Variable EXTRACTION_SPEC_VERSIONConst

    EXTRACTION_SPEC_VERSION: "1.1"

    Spec version: bump on any prompt/schema change; keep the two libs aligned.

    1.1: small models (a 3B model on Ollama, measured) returned "{}" for nearly every real exchange under 1.0, because they read the assistant's reply as part of what to mine. 1.1 frames the conversation, says whose facts to record, sends a JSON schema, and gives two worked examples: one with facts and one where the right answer is none.