PaperQA2

PaperQA2

PaperQA2 by Future House is an AI agent for question-answering over scientific literature. It autonomously retrieves relevant papers, reads full PDFs, and synthesises grounded answers with precise inline citations — achieving human-level performance on the LitQA2 benchmark that evaluates citation accuracy.

Key Features

• Agentic RAG loop: retrieval → reading → evidence synthesis → answer generation
• Exact citation tracking: every claim is traceable to a specific passage and paper
• LitQA2 benchmark leader — outperforms GPT-4 with retrieval on literature QA
• Supports local PDFs, DOI resolution, and Semantic Scholar/PubMed search
• Async Python API; pluggable LLM (OpenAI, Anthropic) and embedding providers

Quick Start

pip install paper-qa
export OPENAI_API_KEY=sk-...

from paperqa import Docs
import asyncio

async def main():
    docs = Docs()
    await docs.aadd("paper1.pdf")
    await docs.aadd("paper2.pdf")
    answer = await docs.aquery("What are the key findings on transformer scaling laws?")
    print(answer.formatted_answer)

asyncio.run(main())

npx ai-supply add paperqa-scientific-literature-qa

Curated mirror of the open-source PaperQA2 (Apache-2.0). Get it from the source.