Perspicacité
A local-first retrieval-augmented research assistant that helps scientists and students search, understand, and organize academic literature across sources such as Semantic Scholar, PubMed, and arXiv, exposed through an MCP server and REST API.
Perspicacité helps researchers search, understand, and organize academic literature, running mostly on the user's own machine so that only the language model inference calls it makes leave the local environment. It searches across Semantic Scholar, OpenAlex, PubMed, arXiv, HAL, and DBLP, ingests papers from PMC XML, arXiv HTML, open-access PDFs, and publisher APIs, and extracts text with PyMuPDF, with an optional layout-aware extractor for tables and figures.
It offers six retrieval-augmented generation modes, from a basic mode through an agentic mode, a literature survey mode, and a contradiction-finding mode, using cheaper models for routing and screening and a stronger model for synthesis. A knowledge base can be built from BibTeX imports, bulk DOIs, local documents, or a Zotero collection, citation graphs can be expanded forward and backward, and the assistant is reachable through an MCP server with 32 tools or a REST API with streaming for asynchronous jobs, recording the source and retrieval trail behind every answer.