PaperQA2
Python package that answers questions from scientific PDFs with cited, agentic retrieval-augmented generation.
PaperQA2 searches PDFs and text files with an LLM agent and answers with in-text citations.
PaperQA2 is a Python package for high-accuracy retrieval-augmented generation over PDFs, text, Office files and code, with a focus on scientific literature. It suits researchers and students comfortable with a command line who want grounded answers with citations.
What it is
You point it at a folder of papers and ask a question with pqa ask. It fetches paper metadata including citation counts and retraction checks, indexes full text, and an LLM agent iteratively refines queries and writes an answer citing the passages used.
Who it's for
It suits researchers, reviewers and lab teams who want a scriptable literature assistant. It does not suit people who want a point-and-click app, and it needs an LLM API key or a local model server.
Key features
Cited answers
Responses carry in-text citations to specific papers and pages.
Metadata awareness
Fetches citation and journal-quality data from providers such as Semantic Scholar, Crossref and Unpaywall, with a retraction check.
Agentic search
A language agent iteratively refines queries and answers instead of a single retrieval pass.
Re-ranking and summarising
Uses LLM-based re-ranking and contextual summarization over retrieved passages.
Any LiteLLM model
Works with all LiteLLM providers, local servers such as llamafile and ollama, and local sentence-transformer embeddings.
Local full-text index
A searchable index of local PDF and text files, with support for code and HTML.
Up and running in 3 steps
Our honest verdict
What's great
- Answers cite sources inside the text, which makes claims checkable.
- Automatic metadata fetch adds citation counts and retraction checks to your papers.
- You can swap in other LLMs and embeddings, including local ones, instead of the OpenAI defaults.
- The quickstart is two commands: pip install paper-qa and pqa ask.
Where it falls short
- By default it calls OpenAI embeddings and models, so you need an API key and pay per use unless you set up a local model.
- Indexing 100+ papers works best with your own Crossref and Semantic Scholar API keys to avoid public rate limits.
- It is a Python library and CLI with no graphical interface.
- The README says you may get different results than the authors' papers, as their paper corpora use APIs and licences that cannot be shared openly.
Pricing & license
PaperQA2 is free under Apache-2.0 and no hosted plan is shown in the repository. Your cost is the LLM and embedding API usage, unless you run local models.
Open remote AI-training roles
767 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Do I need an OpenAI account?
It is the default. You can instead configure any LiteLLM-compatible model, including local servers such as llamafile or ollama.
What file types does it read?
PDFs, text files, Microsoft Office documents and source code files.
Where do I get papers?
The README has a 'Where do I get papers?' section; PaperQA2 indexes the files you give it and fetches metadata for them.
