raglite

What This Skill Does

RAGLite is a high-performance, local-first RAG (Retrieval-Augmented Generation) cache designed for OpenClaw agents. Unlike cloud-based knowledge bases, RAGLite operates entirely on your local machine, ensuring your data remains private and secure. It utilizes a powerful two-stage processing pipeline: first, it distills raw documents (such as PDFs, web pages, or notes) into clean, structured Markdown, which acts as a compressed, human-readable artifact. Second, it indexes this data using a hybrid approach combining Chroma for vector semantic search and ripgrep for lightning-fast keyword matching. This dual-indexing strategy ensures that you receive the most relevant information whether you are searching by conceptual similarity or specific technical terminology. RAGLite is not intended to replace an agent's working memory or long-term chat context, but rather to serve as a durable, queryable library of information that exists outside of the model's primary training data.

Installation

To integrate RAGLite into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/virajsanghvi1/raglite. This will pull the necessary dependencies into your skill registry. For detailed manual configuration, you can also run the provided shell script at ./scripts/install.sh, which provisions a skill-local virtual environment at skills/raglite/.venv and installs the raglite-chromadb package, ensuring that your agent has dedicated, stable access to the required libraries without conflicting with other system-wide packages.

Use Cases

• Personal Knowledge Management: Store and retrieve personal notes, journals, or academic research documents locally without privacy concerns.
• Technical Documentation: Maintain a local index of internal runbooks, API documentation, or complex technical specifications for instant querying.
• Healthcare and Sensitive Records: Securely manage medical history or private legal documents, keeping sensitive data entirely air-gapped from third-party model providers.
• Development Workflow: Quickly digest project documentation and codebase manuals into a queryable cache to assist in rapid development cycles.

Example Prompts

1. "RAGLite: Distill all Markdown files in the ./projects/documentation folder and index them into the 'dev-guide' collection."
2. "Search the 'medical-records' collection for any mentions of 'blood pressure' and summarize the findings."
3. "Run a hybrid query against the 'internal-runbooks' collection for 'how to reset the production database server' and format the output as a step-by-step checklist."

Tips & Limitations

• Distillation is Key: Because RAGLite compresses data into Markdown, verify the distilled output periodically to ensure the model captured the essential nuances of your original source files.
• Security: Always treat source files as untrusted data. RAGLite includes built-in instructions to ignore prompt injections, but maintain standard practices when processing third-party web content.
• Hybrid Search: Leverage both the vector and keyword search capabilities. If your query is vague, rely on Chroma; if you need to find a specific error code or function name, utilize the ripgrep integration.
• Storage: Regularly clean up your output directories if indexing frequent, ephemeral documentation to prevent the Chroma vector database from consuming excessive disk space.