Hey PH 👋 I built rec-format because I kept watching my Anthropic bill climb every time I fed a PDF to Claude. A 100-page research paper? ~160,000 tokens. Just to read it. So I asked a dumb question: what if the file format itself was the problem? PDFs were designed for printers and people. Margins, fonts, embedded images, invisible metadata, none of it means anything to an AI. You're paying for pixels dressed up as text. .rec strips all of that. Same document, same information, same structure but compressed, paginated, and schema-typed so a language model can actually parse it cleanly. ~100 tokens per page instead of ~1,600. On a 100-page doc, that's 150,000 tokens back in your pocket on every single API call. "Why not just convert to Markdown?" — I get this one a lot. Honest answer: token cost inside the context window is nearly identical. The real difference is everything else: Page references. Every block knows which page it came from. Ask the AI "what's on page 9?" and it actually knows. Markdown has no concept of pages. Embedded metadata. Title, language, source file, page count, baked into the file as standard fields, not a frontmatter convention you may or may not follow. Integrity guarantee. SHA-256 checksum so you know the file arrived complete and unmodified. Markdown has nothing like this. Unambiguous schema. No dialects, no parser disagreements. Every block is typed JSON, a footnote is always a footnote, everywhere. Vision model support. Image regions are extracted as base64 PNG and passed directly to Claude or GPT-4V. Markdown just loses them. Is it perfect? No. Tricky layouts still trip it up sometimes. But for anyone feeding research papers, reports, or lecture notes into an AI, it works, and it saves real money. Open format. Open source. Free to use. Happy to answer anything 👇 A.Recai

AI optimized PDF Converter

PDF for AI. AI Native PDF Converter.

PDF for AI. AI Native PDF Converter.