PaperQA2 is a retrieval augmented generation tool for answering questions from scientific documents with citations.
The tool addresses the challenge of extracting accurate, cited answers from large collections of scientific literature. It works by ingesting PDFs, text files, Microsoft Office documents, and source code files, then building a full-text search index with cached embeddings. When a user poses a question, the system retrieves relevant document passages and uses an LLM agent to synthesize answers grounded in those sources, returning results with explicit citations. The approach emphasizes high accuracy on scientific tasks including question answering, summarization, and contradiction detection.
Developers should adopt this tool if they need to build applications that answer questions about scientific literature with verifiable sources. It suits projects where citation accuracy and traceability matter more than speed alone, and where the document collection is primarily scientific papers or technical documents. The README distinguishes the tool from LlamaIndex and LangChain by highlighting its specialized focus on scientific literature and its demonstrated superhuman performance on domain-specific tasks, though it does not provide detailed feature comparisons with those frameworks.
The project maintains active development with regular updates to its core algorithms and capabilities. The tool supports both agentic and manual workflows for document processing, allowing users to choose between automated agent-driven indexing or explicit control over the pipeline. Configuration is flexible, with bundled settings for common scenarios, customizable prompts, and support for multiple embedding models including locally hosted options. The system provides caching mechanisms for embeddings and external vector database integration, reducing redundant computation across queries.
