{"version":"2.1.94","anchor":"llm-based-memory-synthesis","canonical_anchor":"llm-based-memory-synthesis","heading":"LLM-Based Memory Synthesis","tier":null,"area":null,"scope":null,"heads_up":null,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.94\/e\/llm-based-memory-synthesis","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.94","markdown":"### LLM-Based Memory Synthesis\n\n**What**\n\nNew memory retrieval path that uses an LLM to synthesize relevant memories into a single paragraph, replacing the previous raw-file retrieval approach.\n\n**Details**\n\n- When tiny memory mode (`tengu_billiard_aviary`) is active, uses LLM synthesis to produce contextually relevant memory summaries\n- System prompt instructs: \"synthesize information directly relevant to\" the current query\n- Output formatted as JSON with a one-paragraph synthesis\n- Recalled memories shown to users with: \"Recalled from your persistent memory system:\"\n- Synthesis memories are excluded from the standard memory context to avoid duplication\n\n**Evidence**\n\nMemory synthesis (search for `\"Synthesize memory information relevant to\"`, `\"synthesizeRelevantMemories\"`)\n\n- Flag `tengu_billiard_aviary`: Off in both readings (read for one account on one subscription tier against v2.1.94; this account: off, anonymous baseline: off, compiled default: no gate table built for this version) These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release."}