raglite

What This Skill Does

RAGLite is a specialized local-first RAG (Retrieval-Augmented Generation) cache designed to provide your OpenClaw agent with durable, structured access to private documentation. Unlike model memory or short-term chat context, RAGLite creates a persistent knowledge layer on your local machine. The skill operates via a three-stage pipeline: it condenses raw documentation into structured Markdown to eliminate fluff, indexes that data into a Chroma vector database, and enables powerful hybrid search functionality. By combining Chroma's semantic vector similarity with ripgrep's lightning-fast keyword matching, RAGLite ensures you get precise, relevant results for technical documentation, medical records, or personal notes without needing to upload sensitive data to third-party cloud services.

Installation

To integrate RAGLite, first ensure you have Python 3.11+ and the ripgrep binary installed on your system. Navigate to your OpenClaw installation and execute the provided installation script: ./scripts/install.sh. This sets up a dedicated skill-local virtual environment and fetches the necessary libraries from the repository. Ensure your Chroma server is active, typically at http://127.0.0.1:8100, and that your OpenClaw Gateway is configured with the OPENCLAW_GATEWAY_TOKEN environment variable if your specific deployment requires authentication.

Use Cases

RAGLite is ideal for users who manage large amounts of static but sensitive information that a model might not have been trained on. Use it to index complex local codebases, maintaining a searchable repository of internal runbooks for engineering teams. It is perfect for personal productivity, such as organizing large academic research repositories, managing healthcare records locally to maintain HIPAA-adjacent privacy, or indexing years of personal journal entries. Because it creates human-readable, version-controllable Markdown artifacts, it functions as both a database for the agent and an auditable knowledge base for the user.

Example Prompts

1. "RAGLite, index the entire 'Project_Alpha' folder located in my documents directory into a new collection called 'alpha-project' and ensure we skip files already processed."
2. "Search the 'runbooks' collection for the most relevant procedures regarding database connection errors and summarize the top 3 steps identified in the metadata."
3. "Can you perform a hybrid search across my 'notes' collection for 'Q3 financial strategy' and generate a summarized markdown report based on the findings?"

Tips & Limitations

For optimal performance, always ensure your source documents are formatted cleanly before indexing. Use the --skip-existing flag during your runs to save significant processing time and token costs. Note that RAGLite relies on a local Chroma instance; if the service is down, your retrieval queries will fail. If you notice a lack of keyword matches, verify that rg --version returns a valid output in your terminal. Remember that RAGLite is a cache, not an agent's long-term memory store; it excels at static retrieval but should not be used as a primary state-tracking mechanism for active conversations.