raglite

What This Skill Does

RAGLite functions as a local-first retrieval-augmented generation (RAG) cache designed to enhance your OpenClaw agent’s knowledge base. Rather than replacing the model's internal memory, it provides a structured, durable storage layer for private documentation—such as medical records, technical runbooks, or extensive research notes. The skill operates via a three-stage pipeline: it first condenses raw input documents into clean, structured Markdown, then indexes these distilled outputs into a local Chroma collection, and finally enables advanced hybrid retrieval. This retrieval process combines traditional vector-based similarity search (via Chroma) with lightning-fast keyword searches using ripgrep (rg). By distilling content before embedding it, RAGLite reduces noise, cuts down on token costs, and improves the reliability of the information retrieved during agent execution.

Installation

To integrate RAGLite, first ensure you have Python 3.11+ and ripgrep installed on your system. For hybrid search capabilities, install ripgrep using your package manager, such as brew install ripgrep. The skill itself is installed into a local virtual environment managed by OpenClaw. Run the provided installation script: ./scripts/install.sh. This script pulls the source code directly from the official repository at https://github.com/VirajSanghvi1/raglite.git. If you are utilizing the default OpenClaw engine for document condensation, verify that your gateway is accessible and that the OPENCLAW_GATEWAY_TOKEN environment variable is correctly configured in your session.

Use Cases

• Personal Documentation: Store and quickly query local knowledge bases, like Obsidian vaults or personal research notes.
• Technical Operations: Index internal system manuals, runbooks, or deployment documentation to allow the agent to answer "how-to" questions during incident response.
• Compliance & Privacy: Keep sensitive documents off third-party cloud vector databases by utilizing the local-first architecture.
• Document Distillation: Automatically turn unstructured, verbose text files into concise, Markdown-formatted summaries for faster model ingestion and reduced context window usage.

Example Prompts

• "RAGLite, ingest the documentation located in ~/work/project-docs and create a collection named 'dev-guide'."
• "Query the 'dev-guide' collection: 'What is the specific rollback procedure for the production database?'"
• "Condense all my meeting notes in ./notes and index them into the 'personal-archive' collection for future reference."

Tips & Limitations

• Chroma Connectivity: Always ensure your Chroma server is active at the specified URL (default http://127.0.0.1:8100) before running retrieval tasks.
• Optimization: Use the --skip-existing and --skip-indexed flags to optimize your workflow if you are updating documentation incrementally.
• Data Privacy: Because the data is stored locally, it is your responsibility to manage the disk space and accessibility of the ./raglite_out directory.
• External Dependencies: RAGLite requires the presence of rg for keyword search; if missing, hybrid search functionality will fail, reverting to pure vector search. Always verify environment connectivity if you experience empty search results.