All notable changes to MemoryPlus will be documented here.
- Persistent Graphiti Worker
- Added a multiprocessing engine client/worker pair that keeps Graphiti alive between requests.
- Plugin now supports
engine_modewith persistent workers, automatic restarts, and subprocess fallback.
- Queued Ingestion Pipeline
- Introduced an internal ingestion queue with configurable capacity, overflow policies, batching, and exponential retries.
- Graceful shutdown drains pending jobs or logs dropped items during plugin detach.
- Context Search Cache
- Added an LRU-style search cache with TTL and fuzzy matching to avoid redundant Graphiti lookups.
- Cache invalidates whenever new episodes are written to keep results fresh.
- NLP Pipeline Improvements
- Runner now performs lexicon-based emotion detection, topic tagging, lifecycle pruning, and supports new analysis prompts (ResolveEntities, MemoryGate, CustomPrompt).
- Insight generation honors custom prompts while maintaining safe defaults.
- Documentation Refresh
- README rewritten with a product overview, setup guide for Kuzu/Neo4j, configuration breakdown, usage tips, and troubleshooting FAQ.