memory-lancedb-pro

Persistent long-term memory for OpenClaw agents requires LanceDB-backed storage to retain data across sessions. This enables durable memory storage, hybrid retrieval, and self-improvement governance for workflows.

Hybrid retrieval combines vector embeddings with BM25 keyword scoring to improve recall accuracy for agent memory. This mechanism ensures relevant context is matched using both semantic similarity and exact term frequency.

Multi-scope memory isolation for OpenClaw workflows is configured through LanceDB storage boundaries and MCP memory tools. This separates memory contexts across different agents or scopes using explicit validation checks.

LanceDB functions as an embedded memory store without requiring a separate server setup. This embedded architecture allows direct integration for persistent agent memory within the OpenClaw workspace.

Weibull decay lifecycle management applies a statistical decay function to agent memory entries over time. This governance mechanism automatically degrades the relevance of outdated memories to maintain efficient storage and recall.

MCP memory tools supported include memory_recall, memory_store, memory_forget, memory_update, memory_stats, memory_list, and self_improvement functions. These tools manage cross-scope memory with explicit safety and validation checks.