I got tired of PDF editors that break everything you touch

Security goes first, not last. Across the libraries I studied, security was almost always an afterthought. Font files accepted without validation. Remote resources loaded by default. SVG parsed without XXE prevention. These weren’t edge cases — they were architectural gaps. I wrote input sanitization, URL validation, font magic-byte checks, and image decompression limits before writing a single layout function.

Pagination can’t be separate from layout. This was the universal mistake. Every library treats pagination as a post-processing step — run layout, then split across pages. That breaks the moment a container or table crosses a boundary, because the split changes available space, which changes the layout, which moves the split point. My engine resolves them together — same pass, not two separate passes.

Don’t depend on someone else’s runtime. Headless browsers get too heavy for production. Archived projects leave you with a dead dependency. Commercial licenses change terms. I wrote every component — parser, resolver, layout engine, font subsetter, PDF writer — so when something breaks, I fix it. No upstream to wait on.

Keep layout and rendering separate. Several libraries tightly couple position calculation with PDF drawing. That makes adding new layout modes nearly impossible without rewriting the renderer. I kept them in separate layers from day one. Adding Flexbox meant adding a layout class. The PDF writer never changed.

If you create a document in the editor — from a template or from scratch — it works reliably. Add pages, build complex layouts, export. The engine handles large documents well because it was built for that from the start.

Where it gets harder is AI conversion of complex existing PDFs. A clean invoice or a standard contract converts accurately. But the more nested the layout — multi-column sections inside tables inside flex containers, decorative elements overlapping content — the more the AI has to guess, and guesses aren’t always right. I’m improving conversion accuracy continuously, but I want to be upfront: heavily layered documents might need adjustment after import.

That’s exactly why the editor exists. The AI gives you the structure. The editor gives you the tools to make it right. And once you export, that corrected version is preserved — no AI involved next time.