OmniOKF turns messy docs into linked files to cut agent token costs
OmniOKF is a CLI tool that turns PDF, Word, Excel, PowerPoint, HTML, and files into small linked files that an AI agent can search more selectively. Large documents are expensive and slow to place inside a , so the tool splits documents at headings and then uses to refine and label the pieces. Each concept file is about 200 to 500 tokens, so an agent can load only the parts it needs for a question.
The claimed result is a 75% to 95% drop in per query. The tool also creates links between related files with normal relative links, so connected ideas can be followed without a separate . It uses Microsoft’s to read many file types in memory and avoids heavier converters such as Pandoc.
It can also produce Mermaid visualizations to show how the document pieces connect.
Key points
- OmniOKF converts common office and web document formats into linked files.
- It makes small concept files of about 200 to 500 tokens each.
- is used to clean up, refine, and categorize the document pieces.
- The tool claims 75% to 95% lower per query by loading only needed nodes.
- It avoids a and uses links plus Mermaid visualizations instead.