Run a literature review with citations you can check

What this involves

A dependable literature review requires moving past keyword searches and abstract summaries. In this workflow, the research agent queries live bibliographic indexes including Europe PMC, Crossref, OpenAlex, and arXiv directly on its own machine, retrieving genuine publication metadata without requiring any local software installation. This process helps you map established evidence, assess contested mechanisms, and trace experimental lineages across disciplines.

Doing this properly demands that you do not accept abstracts at face value. Abstracts function largely as marketing copy; relying on them risks repeating unverified claims, missing critical caveats, or mistaking pseudoreplication for genuine statistical power. A thorough review inspects the methods within the full text to verify comparator groups, sample sizes, pre-registered clinical endpoints, and whether computational benchmark splits remained uncontaminated.

Every paper cited in the final synthesis must be retrieved directly by the agent rather than recalled from model memory, preventing fabricated references. The agent follows citation graphs to uncover replications and superseding work, extracts exact parameters and uncertainties using PaperQA2 (Crow) and full-text reading tools, and organises findings into established consensus, active disputes, and unaddressed questions.

How it goes

1. 1Construct and dispatch multi-axis queriesThe agent expands queries across biological entities, phenomena, and study designs, running parallel searches across Europe PMC, Crossref, OpenAlex, or arXiv via search_papers alongside author disambiguation using find_author.
2. 2Follow the citation graphUsing lookup_work with references and cited-by parameters, the agent traverses the OpenAlex graph to identify foundational antecedents, replications, and subsequent refutations.
3. 3Retrieve and inspect full-text methodsThe agent locates open-access copies via find_open_access_copy and reads full text using read_paper or PaperQA2 (Crow), extracting sample sizes, blinding, controls, and uncertainties.
4. 4Verify passages and trial registrationsSpecific experimental queries are answered with verbatim extracts via search_document, while clinical claims are verified against registry records using search_clinical_trials and lookup_clinical_trial.
5. 5Synthesise findings with checked citationsThe agent presents the synthesis under established, contested, unknown, and resolution headings, embedding verified reference links and comparative evidence tables created with create_table.

Start it here

Ask in plain language. The agent works on a Linux machine of its own — it installs and runs what the job needs, keeps every file, and reads the results back with the numbers explained. Nothing to set up, and a long run continues after you close the page.

What to watch for

• Never quote a sample size, effect size, dose, or caveat that you have only seen in an abstract.
• Entering an author's name into a plain search query matches titles and abstracts rather than authorship, mixing homonyms instead of isolating their genuine publications.
• Treating multiple culture dishes from a single animal as independent samples causes pseudoreplication, which frequently inflates mechanistic claims.
• High citation counts reflect attention rather than correctness, as retractions and disputed claims are often heavily cited.
• A clinical study's reported primary endpoint must be checked against its pre-registered record, as discrepancies between the two are significant findings in themselves.
• Never average two conflicting results into an artificial compromise; document both positions and examine differences in model systems, doses, or endpoints.

What it runs

Questions