Upload PDFs into a chat. Every answer is drawn from those files and cites the page it came from, so you can check it in a second rather than trust it.
Underlying operating profit was £187.5 million, down 4.0% on 2024 when it was £195.3 million. The margin was 8.7%, against 9.7% the year before.p.59
Each chat holds its own documents: annual reports, board packs, contracts. A chat only ever searches the files you put in it.
The agent decides whether it needs to search. When it does, it queries by meaning and by keyword at once, then a local reranker keeps the best five passages.
Every claim carries a page reference. Hover it and the original text appears, exactly as it sits in the PDF.
Ask for a recommendation and Vouch declines, then shows the figures the documents actually report. The judgement stays with you.
I can tell you what your documents say, but not whether to buy or sell.
Here is what they report instead: sales £2,151.2m, up 6.8%. Underlying profit before tax £171.9m, down 9.4%.p.59
Retrieval carries your tenant and chat identifiers, so one chat cannot reach another's files.
Candidate passages are scored in your own environment before anything is sent to a model.
Files are indexed once into your own vector store and are never used for training.
Each response records the passages it used, with file name and page.
First chat is free. No configuration, no connectors to set up.