Literature notes built on verbatim quotes, not vibes.
Sidenote is a browser extension (Chrome, Edge and Firefox) for working through a literature: open papers on arXiv or as PDFs (scanned archive copies too, via OCR), group them into collections per project, and ask questions across the set. Every answer is pinned to the exact sentence in the source - verified server-side, one click to highlight.
arXiv, journal PDFs & scanned archives · Collections per project
Paper 1 - We find that retrieval improves accuracy only for models under 7B parameters, after which gains vanish.
Paper 2 - In contrast to prior reports, the benefit persists at 70B when the index is domain-matched.
No. Paper 1 reports gains only under 7B parameters 1, while Paper 2 finds the benefit persists at 70B with a domain-matched index 2.
Paraphrase drift poisons a literature review.
A note that says “Paper X found Y” without the exact sentence sends future-you back to re-read the paper - or worse, into print with a claim the paper never quite made. Generic AI summarizers industrialize that failure: fluent summaries, invented findings, no way to tell which is which.
Sidenote inverts it: every claim in every answer is backed by a verbatim quote, verified server-side against the paper before you see it. If the words aren't in the source, the citation is dropped.
How Sidenote works across a literature.
Open papers wherever they are.
arXiv pages, journal PDFs behind your institutional login, uploads dropped onto the side panel. Scanned archive copies are OCR'd so even image-only PDFs become searchable.
Build a collection per project.
Group the papers for a review or chapter into a collection and ask one question across the set. Each citation names the paper it came from.
Quote verbatim, with the line in view.
Click a citation and Sidenote scrolls the PDF to the exact sentence and highlights it. Copy the actual words into your notes - not a paraphrase.
See how collections and verified citations work in detail - or, if you're wrestling with a single dense paper, start with Sidenote for research papers.
The questions a review actually asks.
“Which of these papers measured this effect?”
Ask across a collection and get an answer per paper, each claim cited to the line in the source that reports it.
“Summarize the methods of each paper in this set.”
A per-paper methods rundown with citations, so you can compare designs without re-opening every PDF.
“Find the sentence worth quoting.”
Sidenote surfaces the verbatim passage behind every claim - copy the exact words into your literature notes with the source attached.
“What limitations do the authors themselves admit?”
Pull the caveats and threats-to-validity out of the discussion section, quoted, before a reviewer does it for you.
“Does this scanned 1993 paper actually claim that?”
OCR makes image-only archive PDFs readable, and the citation still lands on the exact line of the scan.
“Is this new preprint relevant to my review?”
A grounded summary of the contribution and its evidence - decide in minutes whether it earns a full read.
Verbatim or nothing.
Sidenote's citations aren't decorative. Before an answer reaches you, every quoted passage is checked server-side against the text actually retrieved from the paper. A quote that doesn't match the source, word for word, is dropped - so a citation you can see is a citation that held up.
That discipline is what makes the answers usable as literature notes: the quote, the paper, and the exact line, captured while you read instead of reconstructed at write-up time.
Share a cited summary with your lab.
Any summary, glossary or cited answer can become a public share link: snapshot-frozen, revocable, and readable by your supervisor or co-authors in a normal browser tab, no account needed. Answers can be written in any of 17 languages.
Common questions from researchers.
Read the literature. Keep the receipts.
Add Sidenote to your browser, open a paper or build a collection, and ask across it. Free tier forever, opt-in 7-day Pro trial, no card required.
arXiv, journal PDFs and uploads · OCR for scanned archives