Write research papers grounded in sources you can check.

Labrix finds the literature, drafts each section citing papers it actually retrieved, and exports in the format of 500+ journals and conferences. You stay the author.

Encrypted · Never used to train AI models · Yours to delete anytime

Work from your sources
Keep papers and citations in view while you draft.
Private workspace
Your research stays private until you choose to share it.
Keep control of the output
Export your work and verify it before you submit.
01 Read the source02 Inspect the citation03 Draft with context
labrix.app/workspace
Draft · Section 2

Self-attention and parallelism

Transformer architectures redefined sequence modeling by replacing recurrence with self-attention[Vaswani, 2017]. This shift enabled fully parallel training and longer effective context, though at a quadratic memory cost. Subsequent work has explored linear-attention variants and sparse approximations to mitigate the bottleneck without sacrificing expressivity.

In this section we examine three families of approaches and the empirical trade-offs each introduces on long-document benchmarks.

Live demo · under a minute

Watch one paper go from search to submission.

Recorded on the live app. The narration is an AI voice. · Watch on YouTube

Check your paper for plagiarism, free.

Checked against millions of papers and the live web. Every match links to its source with a ready citation. Try 500 words without an account; a free account checks 7,500 words a month.

Evidence before answers

What Labrix verifies — and what it doesn’t.

References, not invented authority

Writing suggestions use the papers retrieved for your project. You can inspect the cited work before relying on it.

Checks with clear limits

Our benchmarks test whether references resolve and exported manuscripts compile. They do not claim to prove that every scientific statement is correct.

Your judgment stays central

Labrix supports research and writing. It does not replace author responsibility, peer review, or source verification.

The Research Loop

A unified workflow eliminating the friction between discovery and publication.

01

Discover

Search 250M+ papers across every discipline — or ask a plain-language question and get papers ranked by how well they answer it. Set alerts to get new work emailed to you.

02

Analyze

Extract methods, claims, findings, and gaps from PDFs. Chat with any paper using a citation-grounded RAG pipeline.

03

Write

Draft sections with your library as live context. Suggestions cite the exact source rather than hallucinating one.

04

Publish

One click opens your finished paper in Overleaf — figures and all — or export LaTeX, Word, and slide decks, and coordinate with co-authors in shared workspaces.

Everything your paper needs, connected.

Your literature, evidence, experiments, and manuscript stay in one research workspace.

Writing StudioYou stay the author

Guided paper writing, grounded in your research.

Turn notes, results, and your library into a structured manuscript. Labrix helps you plan each section, draft with evidence, and check what still needs your judgment.

PapersReview articlesGrant proposalsBook chapters
Manuscript plan
Section 2 of 6
1Outline
2Evidence
3Draft
Explain why parallel training matters

Transformer architectures replace recurrence with self-attention, enabling parallel training across each sequence. Vaswani, 2017

Add your own result to complete this paragraph.

Unified Library

PDFs, metadata, and highlights stored securely and accessible globally across all your projects.

LaTeX Export

One-click LaTeX export formatted for 500+ journals and conferences. Automatic BibTeX generation without the formatting headache.

Research Alerts

Track a topic and Labrix watches the literature for you — emailing a clean digest of new papers the moment they’re published. No noise, no manual searching.

  • New papers across every discipline, ranked for relevance
  • Choose your cadence: real-time, daily, or weekly
  • Delivered to your inbox from discovery@labrix.app
Max & Team

Gap Finder

Enter a topic and Labrix reads the recent literature, then hands you a short list of defensible research gaps — each anchored in limitations the authors themselves stated, and issued to you alone.

  • Grounded in what authors actually wrote — never guessed from silence
  • Every gap ships with evidence quotes and a suggested study design
  • Issued exclusively — the same gap is never handed to two researchers
Built in

Check it. Read it in any language.

Originality, machine-writing, and 28-language translation — built into your workspace. Pick a tool on the left and watch it run.

Labrix · Document tools
Interactive preview
sample.txtAnalysis complete

Transformer models capture long-range dependencies through self-attention. Scaling up parameters consistently improves performance across benchmarks.

32 words·1 paragraph·highlighted = matched text
Result
0%
Originality
2 matched passages
Matched sources
arXiv:1706.03762 · Attention Is All You NeedACL 2019 · BERTCrossref · 4% overlap
Interactive preview — sign up to run these on your own papers.

…and a lot more — citations, experiment tracking, notes, presentations, team workspaces.

Help shape Labrix with real research.

Join a small group testing Labrix on real research work. We want candid feedback on what is useful, unclear, and not yet reliable enough.

Data Protection

Your research should stay yours.

Labrix is built for unpublished papers, sensitive drafts, lab notes, and early-stage ideas. Your workspace is private by default, and you decide when content is shared, exported, or processed.

Private by default

Your papers, drafts, notes, and project data stay inside your Labrix workspace. They are not public, indexed, sold, or used to train Labrix or third-party LLM models.

No admin browsing

Labrix is designed so workspace content is not something admins casually inspect. Operational access is limited and reserved for support, security, or legal needs.

You control sharing

Research projects stay private unless you explicitly invite collaborators, create a team workspace, or connect an external integration.

External processing is explicit

Some features send content to LLM providers (such as Anthropic, OpenAI, and Google), only when you use a feature that requires it. Under their API terms your content is not used to train their models, and any provider-side copy kept for abuse monitoring is held only briefly.

Measured, not marketed

The numbers behind every draft.

We ran this benchmark on the full corpus of generated papers in August 2026 — checking that citations resolve and that the exact file you export compiles to a PDF. Nothing here is hand-picked.

98.5%
1,207 / 1,225 DOIs · 284 refs carry no DOI
Citations resolve
Every DOI in the references checked live against the DOI registry.
100%
85 / 85 compile first try
Compiles to PDF
The exact .tex you export, rendered by Tectonic — no manual fixes.
85
full-length manuscripts
Papers benchmarked
Spanning ML, clinical, physics, chemistry, and bioinformatics.
0
of 3,648 citations invented
No made-up citations
Every [n] across 216 generated sections resolved to a paper actually retrieved.
Full corpus · August 2026 · DOIs verified against the live registry · PDFs compiled with Tectonic

Simple, transparent pricing.

One subscription replaces your reference manager, AI writer, and discovery tool.

Free

Write a complete paper — free.

$0
  • 10,000+ citation styles + BibTeX/RIS export
  • Cite-while-you-write + one-click AI rewrites
  • AI-writing checks — 10 / mo
  • Plagiarism checker — 7,500 words / mo
  • PDF annotation + Word & Google Docs add-ins
  • Paper discovery & reference library
  • Import your Word or LaTeX draft · full pre-submission report
  • MCP — connect Claude, Claude Code or Cursor
  • Draft-depth writing · 5 writing projects · 1 GB
  • A full paper draft a month (16 section drafts) · 5 AI edits a day · 15 analyses/mo

Go

For your next submission.

$20/mo
Tax included
  • Everything in Free, plus:
  • Research-depth writing — our strongest configuration on Methods and Results
  • 120 section drafts · 500 AI edits · 60 analyses / mo
  • Plagiarism checker — 25,000 words / mo, every sentence searched on the web, PDF reports
  • 100 writing projects · 10 GB storage
  • Journal submission templates
  • Pre-submission check with AI language read & one-click fixes
  • Gap Finder — 3 evidence-backed gap reports / mo
  • LaTeX, Markdown & DOCX export
  • Email support
Recommended

Max

Deeper evidence and submission checks.

$40/mo
Tax included
  • Everything in Go, plus:
  • Deep writing — our strongest configuration on every section
  • Deeper evidence checks
  • Gap Finder — 10 reports / mo, deeper literature sweep
  • Polished AI presentation decks
  • 300 section drafts · 3,000 AI edits · 200 analyses / mo
  • Plagiarism checker — 50,000 words / mo
  • Unlimited projects · 50 GB storage
  • Advanced venue analysis and specialist submission formats
  • 15+ conference templates

Team

Your whole lab, one workspace.

$200/mo
Tax included
  • 5 seats with Max-level tools
  • 1,000 analyses/mo (200 per seat)
  • Plagiarism checker — 250,000 words / mo (shared)
  • Polished presentation decks (75/mo)
  • Shared workspace + team library
  • Admin controls + usage analytics
  • 250 GB shared storage
  • Dedicated onboarding & support

No credit card for Free · Cancel anytime · Export your data whenever you like.

Enterprise

Built for institutions and multi-lab teams.

Custom governance, security options, onboarding, and support for research organizations that need Labrix at scale.

Custom
Annual contract and security review options
100,000 analyses/mo
6,000 section drafts/mo
Custom seats and workspaces
500 GB+ storage options
SSO / SAML
Data residency options
Dedicated success manager
SLA-backed support

Start with the research you already have.

Create a free workspace, bring your notes and sources, and turn them into a manuscript you can stand behind.