AI for research and academic projects

AI for your research, from first question to final draft.

Find the papers that matter, see when newer work has moved on, understand them in minutes, and keep it all in one project your AI agent can work on, here or in Claude Code.

Start a project free Watch the 1-minute tour

Free to start · Works with Claude Code and Claude Desktop on your own Claude plan

1:12 · captions on

The whole project

One place for the whole research project

Six steps, one project, and an AI that has actually read the papers.

1
Start a project
Your question or assignment brief becomes a project your AI agent can work on.
2
Discover
Search by meaning across full text, and a feed ranked for your project.
3
Spot newer work
Your agent sees when a later paper supersedes, contradicts or extends a claim.
4
Understand
Clean full text. Highlight anything and get it in plain English.
5
Build with evidence
Save passages and notes to the project. Export BibTeX.
6
Work anywhere
The built-in assistant, or Claude Code and Claude Desktop on your own plan.

Why not just search?

Better than scrolling a search engine

Task	The usual way	With PaperVantage
Finding papers	Keyword search and endless scrolling	Search by meaning across full text, plus a feed ranked for your project
Knowing what's current	Hope the top result isn't from 2017	Links newer work that supersedes, contradicts or extends a claim
Understanding	Forty tabs of dense PDFs	Clean full text. Highlight anything, get it in plain English
Keeping track	Bookmarks, copy-paste and a messy .bib	One project: papers, saved passages, notes and BibTeX
Your AI	A chatbot that has never read the papers	An agent that reads them, here or in Claude Code on your plan

Finding papers
The usual way
Keyword search and endless scrolling
With PaperVantage
- Search by meaning across full text, plus a feed ranked for your project
Knowing what's current
The usual way
Hope the top result isn't from 2017
With PaperVantage
- Links newer work that supersedes, contradicts or extends a claim
Understanding
The usual way
Forty tabs of dense PDFs
With PaperVantage
- Clean full text. Highlight anything, get it in plain English
Keeping track
The usual way
Bookmarks, copy-paste and a messy .bib
With PaperVantage
- One project: papers, saved passages, notes and BibTeX
Your AI
The usual way
A chatbot that has never read the papers
With PaperVantage
- An agent that reads them, here or in Claude Code on your plan

Discover

Find papers by meaning, not keywords

Describe what you're after in plain words. PaperVantage searches the full text of papers, not just titles and abstracts, and keeps a feed ranked for your project.

Searches full text by meaning, with keyword matching as a backstop
Import any arXiv paper by link, ID or title
A For You feed that learns from what you save and skip

Search your library and arXiv

Longformer: The Long-Document Transformer

92% matchNew to you

Iz Beltagy, Matthew E. Peters, Arman Cohan

Local windowed attention plus a few global tokens, so cost scales linearly with document length…

cs.CL20204.1k cited

24 2

Linformer: Self-Attention with Linear Complexity

94% matchNew to you

Sinong Wang, Belinda Z. Li, Madian Khabsa, Han Fang, Hao Ma

Self-attention can be approximated by a low-rank matrix, cutting the cost from quadratic to linear…

cs.CL20201.9k cited

24 2

Stay current

Flags in the reader: coming soon

Know when newer work has moved on

PaperVantage compares papers' findings and links a later paper that supersedes, contradicts or extends a claim, so you build on the current best approach instead of a 2017 one.

“Supersedes”, “contradicts” and “extends” links between papers, for your agent today
The citation graph shows what came before and after
Your agent can check a draft's claims against newer work

Classic

Attention Is All You Need

Vaswani et al. · 2017 · cited in your brief

Newer work

2017 claim: attention needs memory that grows with the square of the sequence length. FlashAttention-2 (2023) computes exact attention in linear memory.

supersedes a claim

FlashAttention-2: Faster Attention with Better Parallelism

Tri Dao · 2023

extends

Longformer: The Long-Document Transformer

Beltagy, Peters, Cohan · 2020

Understand

Understand a paper in minutes, not an afternoon

Every paper becomes clean, readable text with its figures, tables and equations. Highlight anything that loses you and the assistant explains it, next to the passage, so you can check it.

Highlight a passage and ask about exactly that text
Answers can draw on the wider literature
Save the passages that matter straight into your project

AI assistant

Ask about this paper

We suspect that for large values of d_k, the dot products grow large in magnitude, pushing the softmax function into regions where it has extremely small gradients.

Why divide by √d_k here?

Big dot products make softmax put almost all its weight on one word, and then learning stalls because the gradients nearly vanish. Dividing by √d_k keeps the scores in a range where softmax stays smooth and the model keeps learning.

£1.85 of £3.00 used · resets 1 Nov

Send

Projects

A project your assistant can work on

Each research question gets a project: its papers, saved passages, notes and a reference list. Ask the assistant to gather papers or save evidence, and it shows you what it will change before it does.

Papers, passages, notes and BibTeX in one place
The assistant asks before it changes your project
An activity log of everything you and your agents did

Project

Efficient attention for long documents

4 research questions · due 12 December

Papers
12
Saved passages
7
Notes
3
references.bib
12 entries

AI assistant

Efficient attention for long documents

Find what improves on Linformer, save the passages on memory cost, and add the papers to my project.

Done

Add 9 papers and 7 passages to Efficient attention for long documents

FlashAttention-2 (2023)
Longformer (2020)
Performer (2021)
…and 6 more

Your agent, your tools, your Claude plan

Bring PaperVantage to the AI you already use

Your projects work wherever your agent does. Connect Claude Code, in VS Code or the terminal, or Claude Desktop, and it runs on your own Claude plan. Or use the assistant built into PaperVantage.

related-work.md · efficient-attention

efficient-attention

CLAUDE.md.claude/skills/notes/related-work.mdreferences.bibYour project folder

## Related work

Early work cut attention's quadratic cost with low-rank projections [@wang2020linformer] and sparse local windows [@beltagy2020longformer]. More recently, IO-aware kernels made exact attention linear in memory [@dao2023flashattention2], which supersedes the original memory claim [@vaswani2017].

Terminal · claude

> Draft the related-work section from my PaperVantage project. Cite the saved passages.

● papervantage · list_evidence("Efficient attention for long documents")

● papervantage · export_bibtex("Efficient attention for long documents")

Wrote notes/related-work.md · 12 citations, all from your project

Built in
The PaperVantage assistant, beside every paper and project. Included with Unlimited, which can also run it on your own model key.
Claude Code, in VS Code or the terminal
Connect it to PaperVantage on your own Claude plan. Your agent searches, reads and cites from your library and projects.
Claude Desktop and more
Claude Desktop, opencode and Hermes Agent connect over MCP. Scripts can use the pv command line and the API.

Connect Claude Code Setup guides for every client →

Pricing

Free for research. £4.99 when your agents do more.

Projects, search, the reader and the citation graph are free. Unlimited adds the built-in assistant and removes the limit on agent tool calls.

Free

£0forever

Everything you need to run a research project.

Projects, semantic search and the For You feed
The full-text reader and the citation graph
Saved passages, notes and BibTeX export
100 agent tool calls every 30 days: enough to try Claude Code

Start a project free

Unlimited

For daily research

£4.99per month

For daily research, with agents doing the legwork.

Everything in Free
The built-in assistant on every paper and project, with a monthly AI allowance
Unlimited tool calls from Claude Code and other agents
Or run the assistant on your own model key

Go Unlimited

FAQ

Questions, answered

Do I need a Claude subscription?: No. The built-in assistant, part of Unlimited, doesn't need one. If you already have Claude Pro or Max, connect PaperVantage to Claude Code or Claude Desktop and use it from there, on your own plan.
How is this different from a search engine or ChatGPT?: A search engine matches keywords and leaves the reading to you. A general chatbot hasn't read the papers. PaperVantage searches the full text, knows which work is newer, and gives your agent the actual passages to cite.
Which papers can I use?: Any paper on arXiv: paste a link or an ID, or search by title. The full text, figures and equations come with it.
What happens to my data?: Your projects, highlights and chats are private to you. We don't sell data, and we ask model providers not to keep or train on it. The Privacy page has the details.

What are you researching?

Type your question or paste your brief. It fills in a new project for you to check and create.

Free. You'll create an account first.

AI for your research, from first question to final draft.

One place for the whole research project

Start a project

Discover

Spot newer work

Understand

Build with evidence

Work anywhere

Better than scrolling a search engine

Finding papers

Knowing what's current

Understanding

Keeping track

Your AI