A research question. A trail of evidence.
Perspicacité is a local-first assistant for finding, reading and organizing scientific literature.
It turns paper collections into searchable knowledge bases and grounds AI answers in the passages it retrieves.
- Open source · Apache-2.0
- Local knowledge bases
- MCP + REST
Evidence flowRetrieve passages, synthesize an answer, and follow its citations back to the papers. A conceptual illustration, not a screenshot.
From discovery to understanding
Bring a scattered literature together.
Relevant papers, full text, bibliographies and research notes often live in different places. Perspicacité connects discovery, reading and synthesis around a collection you can inspect and reuse.
Find papers across databases
With the optional SciLEx integration, search Semantic Scholar, OpenAlex, PubMed, arXiv, HAL and DBLP. Start from a topic, a BibTeX library, DOIs, a Zotero collection or local documents.
Explore content acquisition →Read with the source in view
Acquire structured full text where available, fall back to PDFs, and label abstract-only content. Hybrid retrieval combines BM25 keyword matching with vector similarity to find useful passages.
Understand retrieval →Match the depth to the question
Move from a focused answer to query expansion, iterative research, live discovery, a literature survey, or a comparison of conflicting claims.
Choose a research mode →Keep the evidence trail
Inspect retrieval provenance, follow citation links, and export conversations as RO-Crate bundles. Keep collections connected to Zotero, BibTeX and Obsidian.
See what can be preserved →For researchers and their tools
One collection. More ways to work.
For scientists surveying a field, students building a reading list, and research teams comparing evidence across papers.
A workspace for people
Use the web interface to manage knowledge bases and ask questions. Bring an existing bibliography, explore a new topic, or examine where reported findings agree and disagree.
An evidence service for agents
Connect an MCP client such as Claude Code, Codex or Mimosa-AI to the same research capabilities. Use the REST API or the command line when a scripted workflow fits better.
Start with your own literature
A bibliography is enough to begin.
Install Perspicacité, import a collection, and ask a first question. The source and the contribution guide are available under Apache-2.0.