memory-lancedb-hybrid

What This Skill Does

The memory-lancedb-hybrid skill transforms the standard LanceDB memory plugin into a sophisticated retrieval system by integrating hybrid search capabilities. By combining vector-based semantic search with BM25 full-text search, the agent can achieve significantly higher accuracy in retrieving historical interactions. Vector search captures the intent and conceptual meaning of queries, while BM25 excels at finding specific keywords, names, or technical terms that might get lost in pure embedding-based similarity calculations. The plugin supports flexible reranking strategies, specifically RRF (Reciprocal Rank Fusion) and linear weighting, allowing developers to fine-tune how the final list of memories is presented to the model context. It maintains drop-in compatibility with existing OpenClaw tools like memory_recall and memory_forget, ensuring a seamless upgrade path for users relying on local vector storage.

Installation

1. Ensure the skill is present in your workspace via ClawHub using: clawhub install openclaw/skills/skills/joeykrug/memory-lancedb-hybrid.
2. Navigate to the plugin directory at ~/.openclaw/workspace/skills/memory-lancedb-hybrid/plugin and run npm install --omit=dev to satisfy dependencies.
3. Modify your ~/.openclaw/openclaw.json configuration to include the path in plugins.load.paths.
4. Update the plugins.entries.memory-lancedb section to enable the hybrid object, specifying your reranker preferences (e.g., rrf).
5. Restart the OpenClaw Gateway to force the plugin to index the memory database and initialize the FTS (Full-Text Search) structure.

Use Cases

This skill is ideal for agents that deal with large, specialized datasets where users frequently reference specific product codes, unique identifiers, or technical jargon alongside conceptual questions. It is particularly effective for project-management assistants that need to recall specific meeting notes or technical documentation where semantic search alone might miss the specific terminology used in the request.

Example Prompts

1. "Recall the conversation where we discussed the API error code 503 related to the payment gateway integration."
2. "What was the specific naming convention we decided on for the user-service refactoring task?"
3. "Summarize my recent work notes regarding the Q3 budget report and mention any constraints I recorded."

Tips & Limitations

Note that enabling hybrid search triggers an automatic FTS indexing process upon startup. If your memory store is massive, the first post-installation startup may take slightly longer as the database builds the text index. If FTS index creation fails, the system is designed to degrade gracefully to vector-only search; check the logs if you suspect the index is missing. Always prioritize the rrf reranker unless you have specific requirements for weighting vector versus keyword match ratios.