Pre-submission peer review

Meet Reviewer 2
before they meet
your paper.

Lemma runs your draft past a panel of AI reviewers — surfacing the weak claims, citation gaps and LaTeX errors a referee will pounce on — inside a real editor that compiles to PDF. Fix what they’ll attack while you still can.

Free for 7 days · no card to start · cancel anytime

Example AI peer review · 4 reviewers

“Attention Beyond Tokens”

6.4/10
Borderline
R1 7 AcceptR2 4 Weak rej.R3 6 BorderlineR4 8 Weak acc.
R2

The obvious baseline is missing. You never benchmark against a top-k mixture-of-experts. Every reviewer expects it — it will be the first question asked.

R1

Strong idea, thin evidence. §3 reports a single seed. Add variance over ≥5 seeds and a significance test before this can clear review.

TeX

Undefined reference. \ref{fig:results}doesn’t exist — you likely meant \ref{fig:arch}. Lemma fixes it in one click.

The committee’s verdict — before you hit submit.Illustrative example
The loop

Write. Analyze. Compile.

Everything happens in one workspace — no copy-pasting into a chatbot, no switching tabs to compile.

STEP 01

Write

A real CodeMirror 6 editor with syntax highlighting, autocomplete and a live document outline — or flip to Rich Text and edit the rendered paper directly.

STEP 02

Analyze

Summon the AI companion: simulated peer review, a weaknesses pass, citation checks, related-work finding, and one-click LaTeX fixes — all grounded in your actual document.

STEP 03

Compile

Compile to a polished PDF in the cloud — no TeX install, no setup, no waiting on a build server. Your paper, render-ready in seconds.

The AI companion

Eight ways to make the paper better.

Not a generic chatbot bolted on the side — a panel of focused tools that read your whole project and act on it.

01

Peer Review

Four simulated reviewers debate your draft and return a verdict, scores, and specific objections.

02

Weaknesses Analysis

The harshest read of your argument — gaps, overclaims, and the holes a reviewer will jump on.

03

Fix LaTeX

Detects broken refs, missing packages and syntax errors — and applies the fix with one click.

04

Improvement Ideas

Concrete suggestions to sharpen framing, strengthen experiments and tighten the narrative.

05

Related Work Finder

Surfaces the papers you should be citing — and the ones a reviewer will ask why you didn’t.

06

Citation Check

Flags uncited claims, unused references and mismatched keys between your text and .bib.

07

Review Response Planner

Paste the reviews you got back. It drafts a structured, point-by-point rebuttal plan with you.

08

Free-form Chat

A streaming chat that already has your full document in context. Ask anything, get grounded answers.

The editor

A real editor.
Not a glorified textarea.

CodeMirror 6 under the hood — plus a Rich Text mode that renders headings, math, images and citations inline, so you can edit the paper as it reads.

Rich Text visual modeHeadings, equations (via KaTeX), figures and citations rendered inline as you type.
Outline & autocompleteJump between sections, autocomplete commands, refs and citation keys.
Resizable panels & PDF viewerEditor, AI companion and compiled PDF, side by side, sized how you like.
Drag-drop uploadsDrop figures and data straight into the file tree, backed by cloud storage.
LaTeX
\documentclass{article}
\usepackage{amsmath, graphicx}
\title{Attention Beyond Tokens}
\begin{document}
\maketitle
\section{Introduction}
We propose a sparse routing objective
that minimises cross-entropy \cite{shazeer2017}:
\begin{equation}
  \mathcal{L} = -\sum_i y_i \log \hat{y}_i
\end{equation}
\end{document}
Confidential by default

Your unpublished
work stays yours.

You’re trusting Lemma with research nobody has seen yet. We treat it that way — your drafts are never training data, and Lemma reviews your paper without ever writing it for you.

Encrypted in transit and at rest

Never trains on your drafts

Your manuscripts are not used to train any model. Reviews run on your text and are not retained beyond your project.

Private to you

Projects are visible only to you and the collaborators you explicitly invite — no public listing, no indexing.

Deletable on demand

Remove a project and its entire review history permanently, whenever you want. It’s your work to take back.

A reviewer, not a ghost-writer

Lemma critiques, finds holes and checks citations — it won’t author your paper. The ideas and the prose stay yours to defend.

Pricing

Free for seven days. Then less than a textbook a month.

Every tool unlocked from the first minute of your trial — the full reviewer panel, all 8 analyses, the editor and cloud compile. Pick monthly, or save with annual.

Monthly
$14/mo

Full access, billed month to month.

  • All 8 AI tools & the 4-reviewer panel
  • Unlimited projects & PDF compiles
  • Rich Text editor & cloud storage
  • Cancel anytime
Start 7-day free trial
Save 35%Annual
$9/mo

Billed $108 yearly. Best for regular writers.

  • Everything in Monthly
  • Two months free vs. monthly
  • Priority compilation
  • Early access to new tools
Start 7-day free trial
Lab
Custom

For groups and departments writing together.

  • Everything in Annual
  • Shared projects & roles
  • SSO & admin controls
  • Onboarding & support
Talk to us
Questions

The things researchers ask first.

Is my unpublished work used to train AI models?
No. Your drafts are never used to train any model. Each review runs on your text in isolation and isn’t retained beyond your project — delete the project and it’s gone.
Does Lemma write my paper for me?
No — and that’s deliberate. Lemma reviews: it finds weaknesses, checks citations, fixes LaTeX and rehearses peer review. The arguments and the writing stay yours, so the work is defensible as your own.
Which venues and formats does it support?
Anything written in LaTeX — conference and journal templates, preprints, and theses. Bring your own .tex and .bib; the AI review adapts to the conventions of your field.
Can I move a project in from another editor?
Yes. Upload your .tex and .bib files, or a project .zip, and pick up where you left off. Figures and bibliography come along with it.
What happens when the free trial ends?
Nothing is deleted. Your projects and history stay exactly as they are — you just need an active plan to run new compiles and reviews. Export your work at any time.
How reliable is the AI review, really?
It rehearses the concerns real reviewers raise, grounded in your actual text — missing baselines, thin evidence, uncited claims. Treat it as a dress rehearsal, not a guarantee: it catches the obvious holes while you still have time to close them.
Stop dreading reviewer 2

Get the brutal feedback
before the reviewers do.

Open a project, paste your draft, and let the companion tear it apart — kindly — while there’s still time to fix it.