Weave AI
The manuscript development AI.
General-purpose AI doesn't know your project or your target journal. Weave AI does. It helps you write a stronger paper, from framing your novelty to mock peer review. Trained on 30+ manuscript tasks and grounded in the latest neuroscience.
From start to submission.
Weave AI stays with your manuscript the whole way. It understands your study, helps you write it up, reviews it the way a referee will, and tells you where it stands before you submit. The more of your project you add, the sharper all of it gets.
- Understand
Knows your study.
Brain regions, methods, statistics, findings. The structure of your work, not a summary of it.
See what it finds - Write
Helps you write.
Titles, abstract, significance, figure plan, the skeleton of the argument.
See how it helps - Review
Pushes back.
Mock peer review, claim consistency, bias, reproducibility, field positioning.
See what it checks
Understand10 tasks
Knows your study.
Brain regions, level of analysis, methods, statistics, findings. Weave AI pulls the structure out of your manuscript, so every suggestion and every critique is about your study, not studies in general.
Brain regions
Lists the areas the study investigates - and says plainly that no specific region is named, rather than inferring one from context.
Level of analysis
Places the work at the biological scale it was actually measured at.
- Molecular
- Cellular
- Circuit
- Systems
- Behavioral
Key findings
States the main results the study reports, in the paper's own terms and without deciding yet whether they hold.
Methods
Inventories every experimental technique the study applied, so a methods section becomes a list you can scan instead of prose you have to mine.
Statistical tests
Identifies the tests and analyses the paper reports, named one by one.
Study design
Determines how the work is structured. Design decides what a finding is allowed to mean long before the statistics are run, so this is the reading that everything downstream depends on.
- Randomized
- Observational
- Longitudinal
- Cross-sectional
- Experimental
Study type
Classifies what kind of study this is, which sets the ceiling on what any result in it can support.
- Randomized trial
- Observational cohort
- Animal model
- Meta-analysis
Species
Names the organism the work was done in. It is the first thing a reader needs and one of the first things an abstract tends to leave out.
MeSH terms
Assigns the Medical Subject Headings a biomedical index would file the paper under, so the work is findable in the controlled vocabulary the field actually searches in rather than only in the words you happened to choose.
Keywords
Pulls the terms that actually summarize the study, not the ones that happen to repeat most often. The list you would put on a submission form, or search to find this paper again a year from now.
Write10 tasks
Helps you write.
Titles, abstracts, significance, the figure plan, the skeleton of the argument. The writing you have to do anyway, started from what your study shows. Every suggestion is yours to take, change, or throw out.
Abstract polish
Takes a rough description of what you did, in whatever words you had at the time, and turns it into a formal scientific abstract.
Why it matters
Explains what changes because this work exists - the case you have to make in a cover letter and rarely get to make in the paper.
- Novelty
- Significance
- Impact
Titles
Writes alternative titles that reflect what the study actually found, so the decision becomes a choice between drafts rather than an afternoon.
- Three options
Paper skeleton
Extracts the study's mechanistic structure end to end, so the whole argument fits on one screen.
- Field
- Subfield
- Hypothesis
- Findings
- Conclusion
- Causal strength
- Limitations
Figure plan
Proposes a figure layout from the results: which figures the findings map to, and what belongs in each panel.
- Figures
- Panels
Hypothesis
States the central hypothesis the study tested, which is not always the one the introduction says it set out to test.
Results summary
Summarizes the results section on its own terms: what was found, held apart from what it was later claimed to mean.
IMRaD breakdown
Splits the manuscript into its four sections, summarizes each, and scores how well they hold together - a paper can have four good sections and still not be one argument.
- Introduction
- Methods
- Results
- Discussion
- Coherence score
Plain language
Rewrites the abstract for a reader outside the subfield: a grant panel, a press office, a collaborator in another discipline.
One-line summary
Compresses the whole paper into a single sentence, three ways.
- Three variants
Review11 tasks
Pushes back.
Does what you say match what you found? Weave AI treats your manuscript as a claim to be tested, not a text to be summarized. The reading you'll get from a referee, before you get it from a referee.
Mock peer review
Reads the whole paper the way a careful referee does: what the study claims, what it actually shows, and where the gap between the two sits. It separates the objections that would sink a submission from the ones an author can fix in a week, and it says where it would land.
- Major concerns
- Minor concerns
- Missing controls
- Statistical issues
- Accept to reject
Claim consistency
Walks every conclusion back to the result it rests on. A correlation quietly becoming a cause, a mechanism asserted from an association, a hedge that goes missing between the results and the abstract: it flags the sentence, not just the section.
- Evidence match
- Overclaim detection
- Causal-inflation risk
Alternative mechanisms
Names the explanations the authors did not test, weighs how plausible each one is, and pairs it with the experiment that would rule it out.
- Competing mechanisms
- Plausibility
- Falsification test
Bias and confounds
Design decides what a result can mean, long before the statistics do. This reads the study for the variables that travel with the effect, the population it was measured in, and the distance between who was studied and who the conclusion is written about.
- Design biases
- Confounds
- Sex bias
- Demographic reporting
- Overgeneralization
Methods audit
Inventories every technique the paper leans on and rates how rigorously each one was applied, not how confidently it was described. A method can be state of the art and still be run in a way that cannot carry the claim.
- Imaging
- Electrophysiology
- Genetic tools
- Assays
- Statistics
- Rigor rating
Reproducibility
Asks the only question that matters of a methods section: could another lab run this? It looks for procedural detail that is actually written down rather than implied, for data and code a reader can reach, and for whether randomization and blinding were reported at all.
- Protocol clarity
- Data and code
- Randomization
- Blinding
- 0 to 10 score
Multi-level interpretation
A finding rarely stops at the scale it was measured. This reads the result upward and downward, from molecules to behavior, so the implications land at every level the work touches.
- Molecular
- Cellular
- Circuit
- Systems
- Behavioral
Field positioning
Places the work in its literature: the subfield it speaks to, whether it confirms or unsettles the prevailing view, and the gap it fills.
- Subfield
- Paradigm status
- Literature gap
Closest prior work
Argues the most convincing wrong reading of the study, then takes it apart. If a result can be misread, it is better to see how here than in a review.
- Plausible misreading
- Why it fails
- Correction
Abstract gaps
Reads an abstract for what is not in it, and names the details a reader needs before they can judge the work at all.
- Species and strain
- Sample size
- Dosage
- Remaining gaps
Specific aims
Turns the paper you just read into the grant you have not written yet: the objective it implies, the hypothesis worth testing next, and three aims with the outcome each one would produce.
- Objective
- Central hypothesis
- Three aims
- Expected outcomes
- Impact
Speaks neuroscience.
Most AI writing tools are a general-purpose model with instructions on top. Weave AI is trained on neuroscience research, trained again on the work of developing a manuscript, and connected to a research library that updates every day.
Neuroscience research literature
It read the field before it learned the job, so it knows neuroscience's vocabulary, its methods, and how its arguments are built.
30+ manuscript tasks
Then taught the actual work of developing a manuscript: pulling out a study's structure, helping you write it up, reviewing it like a referee.
New research, added daily
Connected to a research library that updates every day, so it keeps up with the field instead of going out of date.
Trained on the field's research literature. Never on your manuscripts.Security and data
Bring your whole project.
With Research Intelligence, the documents you add become part of what Weave AI knows. Every suggestion and every critique then draws on everything you've added, not just the manuscript on screen.
Security and dataReviewer comments
Past reviews and response letters, so an old objection is answered before it returns.
Journal guidelines
Author instructions and scope, so feedback fits the journal you're aiming at.
Related papers
The studies you build on or argue with, so your work is positioned against the right ones.
Your published work
Earlier papers, so new writing stays consistent with them.
Grant applications
The aims and significance you've already argued for, so the paper delivers on the grant.
Notes and lab records
Meeting notes and protocols, so methods feedback reflects what you did, not just what you wrote.
Earlier drafts
Previous versions and the sections you cut, so nothing dropped creeps back in.
Anything else
A thesis chapter, a preprint, a talk. If it explains the project, add it.
Weave for Word
Write submission-ready manuscripts.
Weave AI works where your manuscript is written. It checks your draft against the target journal's requirements, reviews it the way a referee would, and gives you a readiness score before you submit. Arriving late 2026.
Get early accessYour research stays yours.
Weave AI reads your manuscript and learns your research to do the work you asked for. That is the only thing that happens to either.
Never trained on.
Your papers are not training data. Nothing you upload teaches the model anything.
Never sold or shared.
Not to advertisers, not to data brokers, not to anyone. Your manuscript is not a product we have for sale.
Never read by us.
No one at Weave opens, reviews or samples your manuscript. A human expert reads it only when you ask one to.
Encrypted throughout.
In transit and at rest, from the moment a file reaches us.
Yours to take back.
Delete a manuscript and it is gone. Take your work back out whenever you want.
Questions about Weave AI
What you can use today, what early access gets you, and what it will cost.
What can I use today?
Weave Stream is live now, and Weave AI already works in it: it builds your literature search and surfaces the papers that matter to your project. The manuscript tools on this page are not available yet. They arrive with the products built for them: Weave Atlas in October 2026, Weave for Word in late 2026, and Weave Horizon in early 2027.
What does early access get me?
A place at the front of the line. We will tell you as each manuscript tool opens, and early-access members get in first. The same account works in Weave Stream today, so you do not have to wait to start.
How much will it cost?
Prices are not set yet, and we would rather say so than guess. What is decided: Weave Stream is free to start with today, Weave Atlas will be free forever, and the Weave Horizon beta will be free.
Does it replace peer review?
No, and it is not trying to. What it gives you is the reading your paper would get from a careful referee, early enough that you can still do something about it. Every analysis is a draft for you to judge, argue with and throw out - the decisions stay yours.