cortexreach/memory-lancedb-pro

memory-lancedb-pro is a plugin for OpenClaw (an AI agent framework) that gives agents a more capable long-term memory system. The built-in OpenClaw memory plugin only does basic vector search, finding memories based on semantic similarity. This plugin replaces that with a richer retrieval pipeline. The core feature is hybrid retrieval: it combines vector search (finding memories that mean something similar to the query) with BM25 full-text search (finding memories that share the same keywords), then fuses the two result sets together. After that, a cross-encoder reranking step, using Jina, SiliconFlow, Pinecone, or any compatible endpoint, re-scores the results more precisely. Additional scoring factors include recency boost, time decay, importance weighting, and MMR diversity, a technique that reduces redundant results. The plugin also introduces multi-scope isolation: memories can be scoped to different contexts, globally shared, per-agent, per-project, per-user, or custom-named scopes, so different agents or users don't contaminate each other's memory. A management command-line interface lets you inspect, search, export, delete, and migrate stored memories without going through the agent itself. Noise filtering skips low-quality content like greetings or refusals before storing, and adaptive retrieval detects when a query doesn't need a memory lookup at all, such as simple confirmations or slash commands. The plugin is compatible with embedding providers including OpenAI, Gemini, Jina, and Ollama. It is written in JavaScript and licensed under MIT. The full README is longer than what was provided.