PaperLabAsk a paper questions, and check every answer on the page
PaperLab lets you chat with one research paper at a time. Answers stream in and cite the passages they used, like [C1]: hover a citation to see its page and section, click it and the paper scrolls to that passage and flashes it. Your notes on the paper go to the model too, and answers can cite them like [N1]. Select part of an answer and save it as a note, anchored on the passage it cites and marked AI.
![A question and an AI answer citing [C1] and [C2]; hovering [C1] shows its page and section, a line leads to that passage flashed on the page, and part of the answer is saved as an AI note.](/PaperLab/shots/feature-ask-a-paper-dark.webp)
What you can do
- Answers stream in as they are written, and cite passages like [C1].
- Hover a citation for its page and section; click it to scroll the paper there.
- A short paper is sent whole; a long one is searched first, and only the best passages are sent.
- Your notes are sent too, marked You or AI, and answers can cite them like [N1].
- Press Follow up to ask about an answer; the model sees your questions and passages, never its own earlier answers.
- Pick the model per question, and each answer keeps the model that wrote it.
- Select part of an answer and Save as note: anchored on the cited passage, marked AI.
How it works
Retrieve
The question is matched against the paper’s passages, or the whole short paper is used.
Answer
The model answers from those passages and your notes, and cites each one it uses.
Check
Click a citation: the paper scrolls to the passage and flashes it.
Tips
- Ask narrow questions (“What is the masking rate?”) before broad ones (“Summarise the method”).
- Check every citation behind a claim you will reuse.
- Use a local model for unpublished work; cloud models are tagged “Cloud”.
Questions people ask
Does PaperLab send my whole paper to the model?
Only a short paper is sent whole. For a long one, PaperLab searches it first and sends only the most relevant passages, plus your notes on it.
Why doesn’t a follow-up include the earlier answer?
So a model’s mistake doesn’t compound. A follow-up gets your earlier questions and the passages they used, never the model’s own earlier answers.
Which models can answer?
A local Ollama model, Anthropic with your key, or any OpenAI-compatible server such as OpenAI, OpenRouter, Groq, Mistral, DeepSeek, Gemini, LM Studio, vLLM or llama.cpp.
Works well with
Guides
- How to chat with a PDF locally, without uploading it — Ask papers questions with a local model, and check every answer against the page.
- Can you trust AI with research papers? — Invented citations, and a five-step check for any AI answer.
- Open-source alternative to ChatPDF, Elicit and SciSpace — Online AI paper tools compared with a local, open-source option.
- How to disclose AI use in research — What to record, a disclosure you can adapt, and why checking matters.