Generate a scientific draft from your PDFs with every citation traceable to the page
PaperLab AI drafts six sections from the reference PDFs you upload. Each in text citation links to the source file, the cited page and the excerpt used as evidence. You write and validate the final text.
Free credits · No credit card · Instant access What changed recently
What this solves
PaperLab AI builds a draft from the papers you already selected. You upload the reference PDFs, define the scope (or ask for a suggestion based on the PDFs), and it writes Introduction, Theoretical Framework, Methodological Procedures, Results, Discussion and Conclusion. Each citation points back to a file, a page and an excerpt. The draft is a starting point that you rewrite.
How it works
-
Upload the reference PDFs
Only PDF, several at a time. Each file is read page by page, split into passages and indexed with title, authors and year taken from the document.
-
Define the scope
Write the topic, objective and hypothesis, or ask for a scope suggestion drawn from the uploaded PDFs. From that, PaperLab builds the global context shared by all six sections.
-
Generate, check and rewrite
Read each section against its citations, regenerate any of them with written feedback, edit the text by hand, and save versions you can restore later.
Method
What happens to your file, step by step, and where the result stops being a decision made by a model.
How it works inside
Each PDF is read with PyMuPDF, cleaned, split at sentence boundaries into 1000-token chunks with 200-token overlap (minimum 100 tokens), counted with tiktoken cl100k_base, embedded with text-embedding-3-small at 1536 dimensions and stored in Postgres pgvector. For each of six sections a fixed bilingual query is built from the global context and searched: with three or more PDFs a SQL function partitions by file name, keeping the four best chunks of each file, and anything below 0.5 cosine similarity is discarded.
Where AI is used, and where it is not
Retrieval is deterministic (pgvector similarity, 0.5 floor, per-file quota). Everything after it is a model: gpt-5-mini extracts the global context and the suggested scope, and six agents, also on gpt-5-mini, write the sections and declare their own citations as JSON. The author list in the prompt is built from PDF metadata by code, but nothing verifies afterwards that a cited label or page matches the chunk it claims.
What it accepts
- PDF only, one file per upload, checked by file extension. No DOCX, no pasted text. An upload is capped at 150 MB.
- A free-text prompt describing the intended article. It feeds the global context extraction and every section query.
- Optionally an AI-suggested scope (theme, objective, delimitation, hypothesis) derived from up to 4 chunks per file, 15 chunks total.
- Pro plan reaches projects, scope and single-section regeneration. Generating all six sections at once requires Expert.
- Billing: 30 credits to create a project, 10 to suggest scope, 55 to generate the full draft, 20 to regenerate one section. Reprocessing PDFs inside a paid project costs 0.
What you get back
- Six Markdown sections: Introduction, Theoretical Framework, Methodological Procedures, Results, Discussion, Conclusion.
- A citation list per section: id, reference label, file name, page range and the excerpt (up to 800 characters) the model says supports the paraphrase.
- A list of the open gaps each section declared, plus inline markers ([COMPLETAR], [DETALHAR], [INSERIR DADOS], [NECESSITA REFERÊNCIA ADICIONAL], [PERSONALIZAR]) marking where the author has to write.
- A bibliography labelled ABNT that is in fact reference label plus file name, and a JSON export of the project. There is no DOCX or PDF export.
- Versions: each save stores a snapshot of the sections that can be restored later.
What this tool does not do
- The output is a draft by design. The agent instructions say to produce a starting point with intervention markers, not finished text.
- It cites only the PDFs you uploaded. With no suitable source the agent is instructed to write [NECESSITA REFERÊNCIA ADICIONAL] instead of going to find one.
- No citation is validated after generation. Page numbers and excerpts come from the model, and the code only links each citation to the first retrieved chunk with the same file name.
- Author and year come from PDF metadata or a regex over the first pages. When both fail, the citation label becomes the cleaned-up file name plus the year.
- The six sections are generated in parallel and do not see each other, so cross-section coherence is not enforced. The sequential mode that passes previous sections exists but is not the default.
- Plan
- Pro
- Cost
- 30 credits
What you get
-
Citations you can open
Every author citation carries the source file, the cited page and the excerpt used as evidence, and one click opens the source PDF for you to check that page.
-
It marks what is missing
When no uploaded source supports a claim, the section marks that a reference is still needed. Open gaps found during generation are listed apart, for you to resolve.
-
Your data goes where it is marked
Without primary data, the Results section synthesises evidence from the PDFs and leaves explicit markers where your own findings go, instead of filling the gap for you.
-
Versions and section by section control
Regenerate one section without touching the others, edit any text by hand, save a named version before a big change and restore it if the new one is worse.
Changelog
Every line below is a change that actually shipped, dated by the day it went out.
-
Latest
- The article is only published after credits are debited, so a failed charge saves nothing.
- Partial generation stays free and stays saved: nobody pays for an incomplete article.
- PDF upload and context generation now require an active plan, at no credit cost.
Show 5 earlier updates
-
- A provider failure shows a short message instead of the raw error text on screen.
- Authentication and quota failures now get their own message instead of one catch-all text.
-
- PaperLab opened inside a project starts with that project's scope already filled in.
- Saved results are captured into the project instead of living only in the tool.
-
- Access to PaperLab is checked against your subscription plan on every route.
- The interface is available in English alongside Portuguese.
- An expired trial is shown as expired, with what to do next.
-
- Running out of credits is reported inside the tool instead of failing without explanation.
-
- PaperLab AI launched: it drafts a scientific article section by section from your PDFs.
- A scope suggestion helps you state what the project is about before generating.
- Sections now draw on a wider spread of the PDF instead of repeating the same passages.
- A new project can be created by pressing Enter.
Questions and answers
Does it write my article for me?
No. It produces a draft that you rewrite. The tool itself warns on screen that the generated content is an initial draft to be reviewed, verified and validated before any academic use. The authorship and the responsibility are yours.
Where does the content come from?
From the PDFs you upload. Each section retrieves passages from your files and cites authors from that list. When no source fits, it is instructed to mark that an additional reference is needed. Check the citations anyway.
Can the Results section carry my own data?
Yes, if you provide it. Without primary data the section synthesises the literature and leaves markers where your results go. Any table or number that came from the reference PDFs needs to be replaced by what you actually measured.
Which plan do I need?
PaperLab AI opens on the Pro plan, where you create projects, process PDFs, suggest scope and regenerate sections. Generating the full six section draft in one run is on the Expert plan.
How to cite this tool
Used it in your research? Here is the reference, already filled in with the version you are looking at and today's access date.
ABNT (NBR 6023)
LESSA, P. W. B. PaperLab AI. Versão 2026.08. [S. l.]: Xplore Dados, 2026. Disponível em: https://xploredados.com/en/tool/paperlab-ai. Acesso em: 21 ago. 2026.
APA 7
Lessa, P. W. B. (2026). PaperLab AI (Version 2026.08) [Computer software]. Xplore Dados. https://xploredados.com/en/tool/paperlab-ai
BibTeX
@software{xploredados_paperlab_ai_2026,
author = {Lessa, Patrick Wendell Barbosa},
title = {PaperLab AI},
organization = {Xplore Dados},
version = {2026.08},
year = {2026},
url = {https://xploredados.com/en/tool/paperlab-ai},
urldate = {2026-08-21}
}
Start with the free credits
Create an account and test the tools before deciding on a plan.
Create free account