future-house/paper-qa is a Python package that finds evidence in scientific literature and produces answers with citations. The README introduces the project under the name PaperQA2 and describes it as an agent-style implementation of retrieval-augmented generation (RAG), the approach of combining evidence found through search with the answer. GitHub stars number 9,424, the license is Apache-2.0, and the repository was created on 2023-02-05. The most recent release is v2026.08.12, but the given material contains no changelog for that release, so what changed this time cannot be confirmed.
What this package is actually for becomes clear from the kinds of documents it handles. It processes PDFs and text, Microsoft Office documents, and source code files, and it supports embeddings that reflect document metadata, LLM-based reranking, and contextual summarization. Together with these capabilities, the core flow is an agentic RAG loop that repeatedly refines queries and answers. The agent repeatedly selects paper search, evidence gathering, and answer tools while exploring a research question. Both the CLI and the library can index local documents and produce answers that include citations. The repository description states that the purpose of this package is to answer questions about scientific documents with citations.
That said, the evidence backing any judgment is still thin. The only evaluation score on record is a Moments score of 4.6, and the material contains no measurements that indicate accuracy or reproducibility. Claims about accuracy appear only in the repository description, while the README merely explains what is supported and does not guarantee that performance. The given material also does not clarify what improved in v2026.08.12. There is not enough evidence to verify feature scope and performance together.