memory-lancedb-pro
memory-lancedb-pro is a production-grade long-term memory plugin for OpenClaw that goes well beyond simple vector search.
About
memory-lancedb-pro is a production-grade long-term memory plugin for OpenClaw that goes well beyond simple vector search. It combines BM25 full-text indexing with vector embeddings, applies cross-encoder reranking, and uses a Weibull decay model to naturally fade unimportant memories while keeping frequently-accessed ones sharp. The result is an agent that remembers what actually matters rather than just the most recent things it's seen.
Teams building production OpenClaw agents where memory quality directly affects output — customer-facing assistants, long-running research agents, or any workflow where context drift between sessions is a real problem. Overkill for single-session scripts or simple Q&A bots.
Pros & Cons
Pros
- check Hybrid retrieval (vector + BM25) catches memories that pure embedding search misses
- check Cross-encoder reranking significantly improves precision for recall-sensitive tasks
- check Multi-scope isolation lets you separate memory by agent, user, or project without conflicts
- check Weibull decay model is a principled approach to forgetting — beats arbitrary TTLs
- check Six-category LLM-powered classification (profiles, preferences, entities, events, cases, patterns) structures memory automatically
Cons
- close Multi-stage retrieval pipeline adds latency compared to single-vector lookup
- close Requires configuring embedding providers and optionally a reranker service — more setup than simpler memory plugins
- close BM25 + vector fusion tuning may need experimentation for domain-specific corpora
- close No visual memory browser or inspection UI out of the box
More Memory
Other tools in the same category.
