Open PaperA Quiet Week of Polish
A mid-December run of changes focused on one thing: keeping you in flow while you read, extract, and write.
Mid-December (roughly Dec 15-23) was one of those stretches where the most important work doesn't look like a “feature.”
It looks like a session that doesn't break.
You open a paper, highlight a paragraph, ask a question, pull a couple citations into your notes, and keep going—without the app turning into the main thing you’re thinking about.
That's the product story of the week: Open Paper getting quieter, so your research can be louder.
Reading should feel like reading
The fastest way to lose momentum is a tiny UI fight at the wrong moment.
So we put time into the PDF reading loop: page tracking that behaves, scroll behavior that's less surprising, fewer edge cases when a PDF URL is invalid, and small but meaningful fixes like sticky table headers that stay sticky and preview panels that scroll the way you expect.
None of this changes what Open Paper is.
It changes what it feels like to spend an hour inside it.
When you ask for help, the first answer should be light
A lot of people don't want an AI essay the moment they open a paper. They want a quick orientation:
- What is this paper actually doing?
- Where should I look first?
- What are the claims I might reuse later?
So we nudged the initial summary to be more “minimal but useful,” and cleaned up how citations and extracted references show up in the UI.
The goal isn't to impress you with volume. It's to help you decide whether to read deeply, skim strategically, or move on.
And we polished chat rendering + message fetching so conversations feel continuous—because research chat only works if it feels like a thread you can trust.
Data tables: evidence you can carry forward
One of the most expensive parts of research is the handoff from reading to writing.
You find the key result… and then you lose it in a pile of highlights.
This week we leaned further into the idea that extraction isn't “a spreadsheet feature,” it's a bridge: pull structured claims, numbers, and snippets out of papers and turn them into something you can actually reuse.
That's part of why we thought about calling them extraction tables (instead of “data tables”). The name points to the job: extracting evidence from PDFs in a way that survives the next step—notes, drafts, and citations. But ultimately we stuck with “data tables” to keep the language more familiar and focus on the end result.
We also tightened the UI around creating artifacts so it feels less experimental and more like a natural continuation of reading.
Sharing work without losing context
Research is rarely solo for long.
We made the permissions story more predictable (what a viewer can do, what an owner controls) and improved how the product behaves when you're collaborating: fewer weird states, clearer affordances, and better behavior around project limits.
This is the unglamorous part of collaboration: it's not “invite a teammate,” it's “make sure the teammate can actually be productive once they're here.”
Audio that sounds less like a template
Audio overviews are only useful if they feel like an overview—something you can play while walking, not something that reminds you it came from markdown.
So we focused on the workflow and quality details: more reliable generation, better separation of playback components, and cleaner text-to-speech inputs.
The product bet remains the same: audio is a mode switch that helps you keep up with your own research when you can't be staring at a PDF.
Quiet engineering that keeps the pace up
Some of this week's work was purely about keeping the project healthy:
- clearer local setup docs (DEVELOPMENT.md)
- lint + dead code cleanup
- safer API wrapper / error typing improvements
You don't “use” these changes directly, but you feel them as fewer regressions and faster iteration.
The throughline
Open Paper is at its best when it's a tight loop:
read → highlight → ask → extract → cite → write.
This mid-December stretch was about protecting that loop.
Not by adding more steps.
By removing the little things that steal attention right when you're trying to think.