A PDF looks like a document with paragraphs, headings, and tables. Under the hood it is closer to a drawing instruction set: "place this glyph at coordinates (100, 400)." There is no concept of a sentence, a paragraph, or a heading — just positioned characters and lines. Extracting structured text from a PDF means inferring all that structure from the spatial positions of those characters.
That's the problem this tool solves. It reads the raw PDF, reconstructs the text flow, and produces clean Markdown that an AI model, a note-taking app, or a developer's workflow can work with directly.
Most PDF converters send your file to a server. A server-side process parses it, returns the result, and (often) logs or caches the content along the way. This tool works differently: the PDF parser is compiled to WebAssembly and runs inside your browser. Your file never travels over the network.
WebAssembly (WASM) is a binary format that browsers can execute at near-native speed. It allows heavy processing tasks — like PDF parsing — that would normally require a server to run locally instead. The tool loads the WASM binary once; after that, all conversions happen on your device.
Once the parser has the raw character positions, the converter applies a series of heuristics to reconstruct document structure:
None of these heuristics are perfect. A PDF that was exported from a word processor usually converts cleanly. A PDF that is a scanned image, or one whose creator deliberately scrambled the character order (a common anti-copy technique), will produce noisy output.
Markdown is plain text with minimal syntax. It's readable as-is, and it's the format that large language models (LLMs) are typically fed. If you want to paste a PDF into ChatGPT, Claude, Gemini, or a retrieval-augmented generation (RAG) pipeline, Markdown is the right intermediate format. It also pastes cleanly into Notion, Obsidian, GitHub, and most documentation tools.
Scanned PDFs — where the pages are images rather than text — can't be extracted this way. They require optical character recognition (OCR), which this tool does not perform. If you drop in a scanned document, you'll get empty or garbled output. The same applies to PDFs with copy protection or obfuscated character maps.
Complex layouts — multi-column academic papers, magazines, forms with overlapping elements — often come out with the columns interleaved. The heuristics work well for single-column prose; they're less reliable for dense layouts. For those, the raw text is still useful for AI ingestion even if the visual order is imperfect.
The main use case is preparing PDFs for AI workflows. A research paper, a legal document, a product spec — all of these become much easier for an LLM to process when they're clean Markdown rather than binary PDF. Developers building RAG systems use this step constantly. It's also useful for anyone who wants to copy content from a PDF into a writing tool without the formatting mess that comes from copy-pasting out of Acrobat.