ParancU
Find the evidence in your documents.
ParancU lets you import PDFs, paste text, scan pages or use images, then ask questions and inspect the passages most relevant to what you are looking for.
Rather than hiding retrieval behind a generated answer, ParancU keeps the underlying evidence visible and lets you explore multiple candidate passages.
Multilingual retrieval and an adjustable semantic-lexical balance help connect questions to answer-bearing passages across different kinds of documents.
ParancU Agent Lab
From evidence retrieval to grounded answers.
ParancU Agent Lab builds on ParancU retrieval to fuse evidence across multiple chunks, generate grounded answers, verify their support, and retrieve again when information is missing.
Get ParancU
ParancU 0.1.0 is available on the Apple App Store for iOS 17 or later.
View ParancU on the App StoreExplore the project
- White paper Architecture, retrieval design, evaluation methodology, results and research direction.
- Project overview A concise introduction to the problem, approach, local-first architecture and current evaluation.
- ParancU instructions How to prepare a corpus, configure the OpenAI API key, query documents and inspect retrieved evidence.
- Demo script and test queries Suggested QA, evidence-containment, lexical-anchor and ambiguity tests for exploring retrieval behaviour.
Local-first architecture
Corpus persistence, multilingual E5 embedding, semantic-lexical scoring, ranking and evidence inspection operate on-device. During corpus preparation, ParancU currently uses a constrained OpenAI API step to generate a short summary and one guiding question for each source chunk.