Documentation

Quick start

You do not need an account to run your first search. Type a research topic on the home page and press enter.

• Write a topic, not keywords. “Federated learning for medical imaging privacy” works far better than “federated learning”. The agents use the full phrasing to judge relevance.
• Watch the progress rings. Each agent reports its own status, so you can start reading papers while the literature review is still writing.
• Switch tabs freely. Every agent runs on every search — the tabs are already populated, not separate queries.
• Sign in to keep things. Saving papers, collections, notes, monitoring and exports need an account (email or Google).

How a search works

One query fans out to every agent at once rather than running a pipeline stage by stage. Results stream back as each agent finishes, so the page fills progressively instead of waiting for the slowest source.

1. Your query is dispatched to all ten agents in parallel.
2. Each agent queries its own sources concurrently, deduplicates across them, and ranks what it found.
3. The literature review and gap analysis then read the retrieved papers and write their analysis from that evidence.
4. Results appear per-tab as they arrive; the review streams in sentence by sentence.

Because a source can be slow or briefly unavailable, agents fail independently: one source timing out never blocks the rest of your results.

The ten agents

Every search runs all of them. You are never choosing which to use.

Where the data comes from

Results are fetched live from public scholarly APIs at query time — nothing is served from a stale private index. Sources include:

• Papers — arXiv, Semantic Scholar, OpenAlex, Crossref, PubMed, DBLP, DOAJ, CORE, OpenAIRE and NASA ADS.
• Datasets — Hugging Face, Kaggle, Zenodo, OpenML, Dryad.
• Code — GitHub and connected implementation sources.
• Grants — live NSF and NIH calls.
• Patents — public patent search.
• People — authorship data enriched with ORCID.

A few sources need an optional API key to enable (for example NASA ADS and Kaggle). Where a key is not configured, that source simply self-skips and the others still return.

Literature review & citations

The review is the feature most worth understanding, because it is also where AI tools most often go wrong.

What it does

It reads the papers retrieved for your query and writes a thematic synthesis: 4–6 sections, each introduced by a heading, comparing approaches, noting agreement and conflict, and tracing how the field has moved. It closes with an outlook section.

How citations work

Papers are numbered in your results, and the review refers to them as [n]. Because generation is constrained to the retrieved set, the review cannot cite a paper that was not actually found. The Comparative Analysis table underneath breaks the same papers down by approach, dataset, metric and limitation, and every row links back to the paper it describes.

What it does not do

It does not read full texts — it works from titles, abstracts and metadata. It will not replace reading the papers, and it can still summarise a paper imprecisely. Treat it as a fast, grounded first pass that tells you what to read closely.

Research proposal

From a set of results you can generate a full proposal draft: title, problem statement, objectives, a phased methodology, expected outcomes and an indicative budget. It is grounded in the papers you found, so the problem statement argues from real literature rather than generalities.

Generation only starts when you click the button — it never runs automatically. Export it as a formatted PDF, DOCX or Markdown. It is a scaffold to edit, not a submission-ready document.

PDF reader

On the Reader page you can upload a PDF and ask questions about it — methods, limitations, what a table shows, how it relates to another paper.

Privacy: the file is parsed in memory to extract text, the text is held only long enough to answer your questions, and it is then discarded. The PDF is never stored on disk. Scanned PDFs with no text layer cannot be read, as there is no OCR step.

Citation graph

The Graph page renders a paper and co-authorship network for a topic, so you can see clusters, bridging authors and which work anchors the field. Useful for spotting the group you should be reading — or emailing.

Collections & sharing

Save any paper, dataset, grant or other result to your library, then organise items into collections with notes.

Collections are private by default. Making one public generates a shareable link that anyone can open without an account — useful for sending a reading list to a supervisor, a lab, or a co-author. Public collections are also added to the site’s sitemap, so they can be indexed. Everything else stays private to your account.

Monitoring & email alerts

Follow a topic and IntelliResearch re-runs it on a schedule, then emails you a single digest of what is new — new papers, new calls for papers, new datasets or new funding, depending on what you asked it to watch.

Digests are batched into one email rather than sent per item, and anything already sent is never repeated.

Exporting your work

• References — BibTeX, RIS and EndNote, so results drop straight into Zotero, Mendeley, EndNote or a LaTeX bibliography.
• Proposals — PDF, DOCX or Markdown.
• Collections — share by link, or export the underlying references.

There is no direct reference-manager sync yet; export is the bridge.

Accounts, Pro & billing

Sign in with Google or an email and password. Signing in unlocks saving, collections, notes, monitoring and exports.

Free and Pro

New accounts get 2 free searches with no card required, so you can run the full pipeline before deciding. After that, Pro is a flat $5 per month with unlimited searches and every agent unlocked. See pricing for the full comparison and the current free allowance.

Payment and cancellation

Payments are processed by Lemon Squeezy as merchant of record, which handles tax and invoicing; cards, PayPal, Apple Pay and Google Pay are accepted. Cancel any time from your dashboard — access continues until the end of the period you have already paid for. No annual contract, no setup fee.

Your data

You can export or permanently delete your data from the dashboard at any time. See the Privacy Policy for detail.

Limits & honest caveats

Things worth knowing before you rely on it for something that matters:

• Verify citations before publishing. Output is grounded in real retrieved papers and linked back to them, but AI can still mischaracterise a source. Check anything that reaches a manuscript, thesis or grant application.
• STEM and biomedical coverage is strongest. The open databases behind the search are richest there; humanities coverage is thinner.
• Abstracts, not full texts. Analysis works from metadata and abstracts, except in the PDF reader where you supply the document.
• Upstream sources can be slow or briefly unavailable. Agents degrade independently, so you may occasionally see fewer results in one tab.
• Deadlines change. Conference and grant deadlines are gathered from public sources — always confirm on the official site before relying on a date.
• No OCR. Scanned PDFs without a text layer cannot be read.
• No reference-manager sync yet. Use BibTeX, RIS or EndNote export.

FAQ

Support

Questions, bug reports and feature requests are all welcome — this is actively developed and feedback shapes what gets built next.

• Help and questions: support@intelliresearch.tech
• Anything else: hello@intelliresearch.tech

See also the Privacy Policy and Terms of Service.