PDF to Markdown by actually looking at the page

What it does

MarkPDFdown converts PDFs (and images) into Markdown by rendering pages and sending them to multimodal LLMs through LiteLLM. It aims to preserve headings, lists, tables, and other structural formatting that conventional text-extraction tools often destroy.

The interesting bit

Instead of scraping hidden text layers, it treats each page as a screenshot and asks a vision model to transcribe what it sees. That makes it layout-aware enough to handle formulas and diagrams, though every page consumes API tokens.

Key highlights

• Routes to OpenAI or OpenRouter vision models via LiteLLM
• Supports file-based and pipe-based CLI usage
• Can target specific page ranges or convert standalone images
• Companion desktop GUI app lives in a separate repository
• Modular Python architecture with clean separation of concerns

Caveats

• Requires API keys for external LLM providers; cannot run fully offline
• README touts format preservation but offers no accuracy benchmarks or cost estimates

Verdict

Useful for developers or researchers fed up with broken PDF-to-text conversions and willing to rely on external vision APIs for better fidelity. Skip it if you need fully offline or local processing.