Skip to content

feat: Document as human-readable cache #90

Description

@TomLucidor

Problem or use case

Sometimes the user wants to dig up the exact lines of a document for accurate reference (e.g. mathematical lemmas, legal/financial details), and it would be useful to have a converter that cache documents in the most legible formats. Converting to HTML yields around 15-40% savings, while MarkDown would yield about 35-55%, keeping all the LaTeX equations + graphs/figures.

Proposed solution

Something based on this line of logic but more local pdf2htmlEX/pdf2htmlEX#209

Docling/MinerU/Markitdown/Marker/Unstructured all look interesting as repos as well but they are more general

Area

Memory / knowledge graph

Alternatives considered

Doing them manually seems to be a hassle when it can be automated into a pipeline (get PDF, convert to HTML/MD and cache that for later, agent will re-read text if KG or VectorDB deviates)

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions