kb-retriever

You can search a local knowledge base without loading entire files by using a hierarchical data_structure.md index and progressive retrieval. This approach reads only small windows around grep matches, avoiding full-file loads into context.

Progressive retrieval works by navigating hierarchical indices at each level to target relevant documents, then iteratively refining results across up to five rounds. It reads only small windows around matches and restricts full-file reads to extract information efficiently.

Yes, you need to complete mandatory learning steps via references/*.md files before processing. This pre-processing step ensures the agent selects the right extraction tools, like pdfplumber for PDFs and pandas for Excel, to extract information efficiently.

The best way to extract answers is using hierarchical index navigation with data_structure.md files to target relevant documents. This method uses grep, Read, pdfplumber, and pandas to iteratively refine results across up to five rounds while avoiding loading entire files.

Yes, you can specify a custom directory path for your local knowledge base search. While the default knowledge base location is the knowledge/ directory, the system supports user-specified paths to locate answers in different locations.

Irrelevant results from large documents occur when full-file reads are not restricted. By enforcing progressive retrieval and hierarchical index navigation, the system reads only small windows around matches and iteratively refines results across up to five rounds to avoid this issue.