Free and open source · Built by a nonprofit · macOS, Windows, Linux

Everything your paper needs.
In one window.

The LaTeX editor that writes, compiles, cites and answers reviewers with you. On your own computer.

Download LatexDo See what it does

retrieval-paper — LatexDo ✨ Tidy 📖 Review 🕘 History 12

EXPLORER

RETRIEVAL-PAPER

figures sections method.tex results.texM main.texM references.bib main.pdf results.csvU

OUTLINE

TIMELINE

main.tex references.bib results.tex

                    1\documentclass{article}
                    2\usepackage{amsmath, graphicx}
                    3\usepackage[numbers]{natbib}
                    4
                    5\title{Efficient Retrieval for Long Documents}
                    6\author{Ada Martin \\ Example University}
                    7
                    8\begin{document}
                    9\maketitle
                    10
                    11\section{Introduction}
                    12Long documents remain hard for dense retrievers.
                    13Our method improves recall by $12.4\%$ over
                    14the baseline~\cite{lewis2020} at no extra cost.
                    15
                    16\begin{equation}
                    17  s(q, d) = \frac{q \cdot d}{\|q\| \, \|d\|}
                    18\end{equation}
                    19
                    20% Results follow in Section 2.
                  

main.pdf 1/ 6 100%

Efficient Retrieval for Long Documents

Ada Martin
Example University

October 15, 2026

1 Introduction

Long documents remain hard for dense retrievers. Our method improves recall by 12.4% over the baseline [1] at no extra cost.

s(q, d) = q · d‖q‖ ‖d‖ (1)

main* 1↑ 0 0 Compiled main.tex in 1.4 s Compiling main.tex… Ln 14, Col 42 Spaces: 2 UTF-8 LaTeX pdfLaTeX

A paper is more than typing LaTeX. It is builds, errors, citations and reviews. LatexDo puts all of it next to your manuscript.

Source and PDF

Click a sentence. Land on its source.

Source and PDF sit side by side. Click anywhere in the PDF to jump to the line that produced it, or go the other way. Edit as code or visually, in the same file.

retrieval-paper — LatexDo

main.tex

                      11\section{Introduction}
                      12Long documents remain hard
                      13for dense retrievers.
                      14Our method improves recall
                      15by $12.4\%$ over the baseline.
                      16
                      17\begin{figure}[t]
                      18  \includegraphics{recall}
                      19  \caption{Recall by length.}
                      20\end{figure}
                    

main.pdf1/ 6

1 Introduction

Long documents remain hard for dense retrievers. Our method improves recall by 12.4% over the baseline.

Figure 1: Recall by length.

main PDF ⇄ source sync pdfLaTeX · SyncTeX

Problems

LaTeX errors, in plain language.

No more thousand-line logs. LatexDo tells you what went wrong, where, and how to fix it, then takes you to the line. It also checks structure, citations and the PDF before you submit.

retrieval-paper — LatexDo ²

main.tex

                    39\begin{figure}[t]
                    40  \centering
                    41  \includegraphics{figures/results.pdf}
                    42  \caption{Recall on long inputs.}
                    43\end{figure}
                    44As shown by \citet{devlin2019}, ...
                  

PROBLEMS 2 OUTPUT TERMINAL CHECK ANALYSIS

main.tex retrieval-paper

Figure file not found: figures/results.pdf Ln 41

What happened. LaTeX could not find figures/results.pdf, so the build stopped. How to fix it. The folder has figures/result.pdf. Check the file name. Go to line 41Use result.pdfShow log

Citation “devlin2019” is not in references.bib Ln 44

main* 1 1 Build stopped: 1 error Ln 41, Col 20

AI, on your terms

An assistant that never leaves your laptop.

Run a model locally, connect Ollama, or bring your own provider. Nothing in your paper changes until you approve it.

retrieval-paper — LatexDo

results.tex

28\section{Results} 29\label{sec:results} 30 31Our approach is really much better than the others in a lot of cases. 31Our approach outperforms prior methods on four of five benchmarks (Table 2). Proposed by AI Approve Reject 32 33\begin{table}[t] 34 \caption{Recall@10 on five benchmarks.}

AI Assistant

Local · Qwen2.5 7BApproval required

Make line 31 precise. Use the numbers in @results.csv

Benchmarks 1, 2, 3 and 5 improve; benchmark 4 is flat. I rewrote the sentence to say exactly that.

Edit results.tex+1 −1 Waiting for your approvalApplied · you approved it

Ask the AI anything

main* Model running on this computer · offline Ln 31, Col 1

Local modelsDownload once, then work with no internet.

Your files as contextType @ to attach a project file to the question.

Talk instead of typingDictation with speech-to-text that runs on your machine.

Citations

See how your references connect.

Search Crossref and OpenAlex without leaving the editor and insert the citation. Then open the knowledge graph to see how your bibliography relates to your paper.

retrieval-paper — LatexDo

main.tex Knowledge Graph

In references.bib · 5 Suggested · 4

lewis2020

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

Authors: Lewis, Perez, Piktus, et al.
Venue: NeurIPS 2020
In paper: Cited 3 times, main.tex
Linked: 2 suggested papers

Insert \citeOpen DOI

dense passage retrievalCrossref · OpenAlex

Dense Passage Retrieval for Open-Domain Question AnsweringKarpukhin et al. · EMNLP 2020 · 2,431 citations

Approximate Nearest Neighbor Negative Contrastive LearningXiong et al. · ICLR 2021

↵ Insert \cite and add to references.bib↑↓

Review and rebuttal

From reviewer comment to response letter.

Annotate any PDF in Reviewer mode. In Rebuttal mode, answer each reviewer point by point, with the manuscript beside you and a response template ready.

retrieval-paper — LatexDo

main.tex Rebuttal

REVIEWER 1

Comment 1 AnsweredClarify the training data.

Comment 2 AnsweredAdd a latency comparison.

REVIEWER 2

Comment 1 DraftingThe claim in Section 3 is not…

Comment 2 OpenMissing ablation on chunk size.

REVIEWER 3

Comment 1 OpenTypo in Equation 4.

Reviewer 2 · Comment 1

“The claim in Section 3 is not supported by the reported numbers.”

Response

We agree. Section 3 now reports exact Recall@10 on all five benchmarks, and the claim is limited to the four where our method improves

main.tex:42 · Section 3 Table 2

Rebuttal 2 of 5 comments answered response-letter.tex

The rest of the toolbox is built in.

Bring your paper with you.

Import Word, Markdown, a searchable PDF or an Overleaf project and keep working in LaTeX.

Undo anything.

Built-in Git, diffs, blame and a local history of every file.

Figures and tables, handled.

A TikZ canvas, draw.io to TikZ conversion and a table formatter.

Dictate a paragraph.

Voice dictation, transcribed on your own machine.

Start from a template.

Templates for articles, research papers, lecture notes, Beamer slides and letters.

Your files stay your files.

Projects are ordinary folders on your disk, compiled on your machine. No account, no upload, no lock-in.

Research

Look inside the writing loop.

See how a build becomes a useful error and how AI acts on a project. The compiler, AI harness and knowledge graph, in motion.

Explore the research

main.tex + references.bib CompilerAI harnessKnowledge graph A page. A fix. A connection.

A nonprofit association

Built for researchers, not for profit.

LatexDo is a French nonprofit association (law of 1901). Our mission is to make scientific and technical writing with LaTeX more accessible, reliable and transparent, for students, researchers, educators and reviewers.

Free writing tools A desktop app with no subscription and no paid tier, open source under the MIT license. Our programs Plain-language guides Documentation that explains LaTeX errors and setup to people who are just starting. Fix LaTeX errors Public accounts Funded by donations and volunteers. Every expense of the association is published. See the expenses

About the association Volunteer

€0Free for everyone, built by a nonprofit

MITOpen source, developed in public

3Platforms: macOS, Windows and Linux

12,500+Automated tests run on every build

Questions, answered.

Is LatexDo free?

Yes. LatexDo is free to download and use, and its source code is open under the MIT license. It is built by a French nonprofit association and funded by donations.

Does LatexDo work offline?

Yes. Your projects are ordinary folders on your disk and are compiled on your own machine. A local AI model also works with no internet once it has been downloaded. Cloud AI providers are optional.

Which operating systems are supported?

LatexDo runs on macOS (Apple Silicon and Intel), Windows (64-bit) and Linux (64-bit AppImage).

Do I need LaTeX installed?

LatexDo compiles with the TeX distribution on your computer: MacTeX on macOS, MiKTeX or TeX Live on Windows, and TeX Live on Linux.

Can I move my paper from Overleaf or Word?

Yes. You can import an Overleaf project, a Word document, a Markdown file or a searchable PDF and keep working in LaTeX.

Is my paper uploaded anywhere?

No. LatexDo needs no account and your files stay on your computer. If you choose to connect a cloud AI provider, only the text you send to the assistant goes to that provider.

Write your next paper in LatexDo.

Free to download. Open the folder you are already working in.

Download LatexDo View the source

Everything your paper needs. In one window.

Click a sentence. Land on its source.

LaTeX errors, in plain language.

An assistant that never leaves your laptop.

See how your references connect.

From reviewer comment to response letter.

The rest of the toolbox is built in.

Bring your paper with you.

Undo anything.

Figures and tables, handled.

Dictate a paragraph.

Start from a template.

Your files stay your files.

Look inside the writing loop.

Built for researchers, not for profit.

Questions, answered.

Write your next paper in LatexDo.

Everything your paper needs.
In one window.