QPaper keeps everything you expect from a good reader — faithful typography, annotations, real page numbers — and quietly adds an AI assistant that has already read the paper with you.
Reading research papers is an act of negotiation between author and reader. The reader skims, retreats, cross-references, annotates — a non-linear movement poorly served by linear summarisation tools [4, 11]. Modern assistants collapse this process into a single query-response loop, discarding the reader's slow, embodied context.
In this paper we argue for a second mode: an assistant that sits beside the page, citing the same text the reader sees, preserving pagination and figure numbers as shared referents. We ground this in a reader study and …
Prior retrieval-augmented generation systems [2, 7] optimise for factual recall on short-answer benchmarks. We instead evaluate comprehension depth and annotation behaviour in open-ended reading sessions.
Annotation studies [9] show readers build meaning incrementally through marginal notes. Our system preserves this process by surfacing citations tied to the reader's current page view, rather than a decontextualised summary.
Participants were assigned one of three conditions: unaided reading, a summarisation-only assistant, or QPaper's side-by-side mode. Sessions lasted 45 min; post-task interviews were coded for comprehension markers (§4.2).
| Condition | Comprehension | Ann. Reuse |
|---|---|---|
| Unaided | 61.2 ± 8.4 | — |
| Summary-only | 67.8 ± 7.1 | 12% |
| QPaper | 79.3 ± 6.2 | 41% |
The controls you already know, with a quiet layer of intelligence on top — grounded in the page you're actually looking at.
Scroll, zoom, search, bookmark, select text, annotate with ink and highlights, and export — everything you'd expect from a PDF viewer, all in the browser.
Every answer links back to the exact paragraph and page it came from. Select a passage and ask about it directly — no re-uploading, no re-pasting.
Your library lives on your machine. Indexing happens locally; only the passages you ask about are sent to the model — and you can see exactly which ones.
Open a local file in the web app, or turn any PDF you encounter in Chrome into a QPaper session with one click.
A full reading environment in your browser. Drag in a PDF, read, annotate, and ask the AI — everything stays on your device.
Every PDF you open in Chrome — from arXiv, a journal, or a colleague's Dropbox — becomes a QPaper session. Same reader, same chat, nothing to upload.