Agent Studio
A live public workspace where agents answer from your own PDFs and cite the page.
A Flask and LangGraph application with local embeddings and retrieval over the visitor's documents. Answers use up to six passages, and the calculator and web search tools run only when the question calls for them. Grounding and retrieval evaluation suites gate every deploy.
Annotation: what gates a deploy
Four evaluations gate every deploy: grounding, classification, citation, and retrieval hit rate.
Citations appear in the interface, so a reader can open the page an answer came from. Embeddings run locally. The calculator and web search tools are used only when a question calls for them.