Preserving Excel cell styles (colors, fonts, alignment) in extraction output

[guigonzalezz]

Hey! first of all, great library, the rust core makes it blazing fast

I'm using Kreuzberg to extract content from .xlsx files and feed the result to an AI agent/LLM for processing. The issue I'm running into is that the extracted output (both with the default markdown and with output_format='html' contains only the raw cell values, all styling information is stripped:

• Cell background colors
• Font colors, bold, italic, underline
• Text alignment
• Merged cells

For my use case this is a problem because the styling carries semantic meaning (e.g. red rows = critical items, green = approved, bold headers, color-coded categories). When the LLM receives plain markdown tables, it loses that context entirely

Here's what I tested:

from kreuzberg import ExtractionConfig, extract_bytes

# Default markdown output
result_md = await extract_bytes(xlsx_bytes, mime_type)
# → produces plain markdown table, no style info

# HTML output format
config = ExtractionConfig(output_format="html")
result_html = await extract_bytes(xlsx_bytes, mime_type, config=config)
# → wraps the same markdown in <pre>, still no style info

# Tables field
result_md.tables[0].cells
# → list[list[str]], plain text only

I also noticed the docs mention "cell metadata" as a capability for spreadsheets. Is there a way to access this metadata (styles, colors, formatting) through the current API that I might be missing?

Ideally, I'd love one of these:

• An option to produce HTML output with inline styles reflecting the original Excel formatting
• Access to cell-level metadata (background color, font properties) in the ExtractedTable structure
• Any other approach within Kreuzberg to retain formatting information

Currently I'm falling back to openpyxl to read the style objects and build styled HTML manually, but I'd prefer to keep Kreuzberg as the single extraction layer if possible.

Thanks for any guidance!

[Goldziher]

Thank you for this topic. I made an issue from it -- I think its a fine request, and there is no issue adding this in the future. i cant commit to when, but i think it will not be too long.

[guigonzalezz]

Thanks for taking the time to look into this and for creating the issue, I really appreciate it.

I’ll keep an eye on it and look forward to any updates. Thanks again!