AI research paper search

Paper search that reaches behind your library login

Your AI sweeps arXiv, Scholar, and PubMed, opens the full PDFs through your library access, and returns an annotated reading list.

[ 01 ]How it works

From one prompt to a sourced reading list

Connect the AI you already use, whether ChatGPT, Claude, or Gemini, to your browser. From one prompt it runs four steps, and reads the full text through the access you already have.

  1. 1

    Point your AI at your own browser

    Link ChatGPT, Claude, or Gemini over MCP. It gains the same Chrome you read in, with your arXiv tabs, your Google Scholar, and your university library proxy already signed in.

  2. 2

    Describe the question, not just keywords

    Give it the topic, the venues, and your filters, like last 2 years, most-cited, reviews only. It searches arXiv, Scholar, PubMed, and OpenReview live, never a stale index.

  3. 3

    It opens the full texts through your access

    For papers behind a publisher wall it uses the institutional or personal access you already have, opens the PDF, and actually reads the methods and results.

    Abstract-only search never gets past the paywall page.

  4. 4

    It hands back an annotated reading list

    Each paper comes with a takeaway, the method, why it’s relevant, and a direct link to the page or PDF it opened, so every line is traceable to a source.

[ 02 ]Read, not just abstracts

Every paper is sourced. Saving stays yours.

A reading list is only useful if you can trust it. Each item links to the page or PDF the agent actually opened, and writing it to your notes or a shared doc works like every consequential Actionbook action: you see exactly what it wants to do before it commits.

  • Linked to the source: Title, authors, venue, and DOI come from the page it read, so you can click through to confirm any claim.
  • You approve the save: Writing the list to your notes surfaces for your approval first, in plain language.
  • Watch it work: Every search and PDF opens in your own Chrome tab, so you can take over at any point.
[ 03 ]Who it's for

Built for how researchers actually read

One workflow, tuned to whatever you’re reviewing this week.

PhD students

Turn a weekend of a literature review into one prompt: it sweeps the venues, opens the gated reviews through your uni login, and returns an annotated reading list.

Annotated reading list
Diffusion priors for de novo protein designNeurIPS 2024Conditions on backbone geometry; 3 fold families.
A systematic review of CRISPR base-editing safetyCell Reports 2025 · your accessAggregates 41 trials; off-target rates by editor class.
Retrieval-augmented generation: an evaluation surveyACL 2025Compares 12 eval setups; reused method table.

R&D engineers

Track a fast-moving field. Ask for the most-cited papers since last year, get each method summarized with a link, and stay current without reading every PDF yourself.

Clinicians

Check the evidence behind a decision. It pulls systematic reviews from PubMed and your subscription journals, with each study’s sample size and conclusion sourced.

Full text via your own access

Most citable reviews live behind a publisher wall. It opens each one through the library proxy or subscription you already log into, reading the methods and results, never bypassing a paywall.

Writers & analysts

Fact-check a claim against the primary source instead of a secondhand summary. It opens the actual paper and quotes the line it read, with the link to confirm.

[ 04 ]Copy a prompt

Prompts to start your literature review

Copy a prompt, paste it into your AI, and Actionbook runs it in your own browser session.

Annotated reading list

Build me an annotated reading list on diffusion models for protein design from the last 2 years. Pull from arXiv and OpenReview, open the full papers, and give each a one-line takeaway, the method, and a link.

Open DOIs via my library

Open these 5 DOIs through my university library access, read the full text of each, and summarize the methods and key findings with the sample size where reported.

Trace the citations

On Google Scholar, find who cited this paper in 2025, open the three most relevant citing papers in full, and tell me what each found and whether it agrees with the original.

Compare eval setups

Compare the evaluation setups of these 3 papers on retrieval-augmented generation. Open each PDF, pull the datasets, metrics, and baselines, and put them in one table with links.

PubMed evidence check

Search PubMed for systematic reviews on GLP-1 agonists and cardiovascular outcomes from the last 18 months, then give me a sourced summary with each study’s sample size and conclusion.

Track a field

Find the most-cited 2025 papers on mechanistic interpretability across arXiv and OpenReview, read the abstracts and the methods sections you can open, and rank them by relevance to my work.

Frequently asked questions about AI research paper search

What researchers ask before letting an AI near their library access and reading lists.

What do you want your agent to do today?