Read papers in batches, extract evidence, and organize a source-grounded literature review you can keep editing.
The slow part of a literature review is never the writing. It is reading fifty to a hundred papers, tracking who claimed what, who contradicted whom, and which findings genuinely disagree, then reorganizing all of it into a line of argument that actually holds together. Skip that and you end up with a pile of abstracts rather than a review.
The review agent takes over that stage. You hand it the papers; it reads each one, extracts claims and evidence, clusters them by theme, and gives you an outline to approve before writing a single section. Once you approve it, the agent drafts section by section with citation numbering maintained throughout. What you get back is a long-form document you can keep editing, not a chat reply.
Upload PDFs in bulk. Each paper is read in full rather than sampled, and its research question, method, data, and conclusion are extracted. Where papers genuinely disagree, the conflict is flagged instead of averaged away.
It does not dump a finished essay on you. It groups the literature by theme, proposes an outline with a clear narrative spine, and waits while you reorder sections, add themes, or drop them.
In-text citations and the reference list share one numbering scheme. Add or remove a paper mid-draft and the numbers renumber, so you never end up citing [12] against a ten-item list.
The output is a document you keep working on. Ask for one section to be rewritten, reframed, or extended with newly found papers without regenerating the whole review.
Here are 80 papers on large language models in educational assessment. Turn this into my thesis review chapter, focused on how the methods have shifted over the past few years.
First an outline organized as early rule-based work, the statistical learning period, the LLM era, and open disputes; after approval, a fully cited chapter draft.
I am writing a review on gut microbiota and metabolic disease. Start by showing me which findings from the last five years actually contradict each other.
A list of genuine conflicts with supporting papers on each side, the sample differences involved, and plausible explanations, ready to become the disputes-and-outlook section.
Batches in the hundreds. With large sets it reads in stages and retains a structured summary of every paper, so earlier reading is not lost as the count grows.
The review is built from the papers you supply plus real sources it retrieves, and every citation points at a specific one. When a claim has no support in the available material, it says the material is insufficient rather than inventing a plausible-looking citation.
You get a structurally complete, fully cited draft that removes the synthesis and drafting time. The scholarly judgement and final verification remain yours, so read it through before submitting anywhere.
Yes. Chinese and English sources can be mixed in one project, and the output language is whatever you specify regardless of the source language.