memory-lancedb-pro-openclaw

What This Skill Does

The memory-lancedb-pro plugin transforms standard OpenClaw agents into long-term learning entities. By integrating a local LanceDB vector database, this skill allows your agent to store, index, and retrieve user preferences, project-specific context, and historical decisions. It utilizes a sophisticated hybrid retrieval system, combining dense vector embeddings with BM25 full-text search to ensure that retrieved data is both semantically relevant and keyword-accurate. With integrated cross-encoder reranking, the skill optimizes retrieval quality, ensuring the most pertinent memories are surfaced before the agent generates a response. It also features Weibull decay-based forgetting, allowing the agent to naturally prioritize recent interactions while allowing less relevant historical data to fade over time.

Installation

The recommended installation method is the one-click setup script, which automatically configures provider presets for Jina, OpenAI, and local LLMs. Run the command curl -fsSL https://raw.githubusercontent.com/CortexReach/toolbox/main/memory-lancedb-pro-setup/setup-memory.sh -o setup-memory.sh followed by bash setup-memory.sh to begin. For advanced users or automated pipelines, you may also install via the OpenClaw CLI using openclaw plugins install memory-lancedb-pro@beta or standard npm packages. Please note that if using npm, you must manually specify the absolute path to the plugin directory within your openclaw.json file under plugins.load.paths to avoid initialization failures.

Use Cases

• Project Management: Track ongoing project requirements and technical constraints across multiple development sessions.
• Personalized Assistance: Remember specific coding styles, preferred response formats, or user-defined shorthand commands.
• Contextual Knowledge Retention: Automatically extract actionable insights from long conversations without needing a database administrator.
• Multi-tenant Isolation: Segment memories by user, project, or agent scope to prevent cross-contamination of private data.

Example Prompts

1. "What was the preferred naming convention we decided on for the authentication module last week?"
2. "Please remember that I prefer all code snippets in Python to include type hints and docstrings."
3. "Summarize the last three meetings we had regarding the project architectural roadmap."

Tips & Limitations

For optimal performance, ensure your embedding model is consistent across all indexed data. Avoid using sessionMemory for very high-volume agents as it may impact retrieval latency. The smartExtraction feature is highly effective but requires a sufficient number of message turns to trigger reliably; set extractMinMessages to at least 2 or 3 for the best balance between speed and quality.