Why I'm making a PDF viewer — Part 2

Update: This project has become ultracontext.co.

I picked this project back up recently. The original idea — a better PDF reader — still made sense to me, but the scope changed.

I'd been using Claude Code a lot, and having an agent that can scaffold and iterate on an app this fast made me rethink what I was actually building. The PDF viewer was always about making it easier to work with documents. But the thing I actually wanted was a workspace where I can throw a bunch of sources at AI agents and have them do the processing for me.

When I'm researching something, I open papers one by one, skim them, copy bits into a chat, and try to keep track of everything in my head. That's most of the work — just processing information. The actual thinking is a small fraction of the time I spend.

So the PDF viewer became ultracontext.co — a workspace designed to scale human thinking. An infinite canvas where you can drop PDFs, videos, web pages, whatever — and agents handle the reading, summarizing, and cross-referencing. You just make the decisions that matter.

The PDF features from Part 1 — splitting, combining, fast switching, good search — are still part of it. They're just one layer now.