Answers you can check, page by page
Kelriva Cite turns a folder of contracts, reports and policies into a workspace you can question in plain English. Every answer names the page it came from. Run it on your own hardware and the documents never leave your building.
Illustration using a public FCA document, FG21/1.
Six real answers, each one checkable
These are recorded answers from Kelriva Cite in cloud chat, word for word, with all five documents loaded at once. Open the cited page in the publisher's own PDF and check it yourself. The last one asks for something the documents do not contain.
In the FCA's Consumer Duty guidance FG22/5, what are the four outcomes of the Consumer Duty?
FG22/5 names four outcomes. They are a set of rules and guidance that set out more detailed expectations for how firms behave in four key parts of the relationship between firms and consumers:
- The governance of products and services
- Price and value
- Consumer understanding
- Consumer support
The guidance explains that these outcomes are one of three parts of the Consumer Duty. The other two are:
- the Consumer Principle, which sets the overall standard of behaviour expected from firms, and
- the "cross-cutting rules", which set three overarching requirements and help firms interpret the four outcomes.
(fca_fg22-5_consumer_duty.pdf, page 3, para 1.3)
The guidance also says that meeting the four outcomes is not everything the Principle or the cross-cutting rules require. The cross-cutting rules set standards of conduct in areas the outcomes don't explicitly cover (page 29, para 5.5).
- 80
- questions on 5 public documents (581 pages)
- 79–80
- answered correctly in cloud chat, across three models
- 9 of 9
- unanswerable questions refused, nothing invented
- 97–99%
- of answers cite a page that holds the answer
Our own test, run on 21 October 2026: FCA guidance FG22/5 and FG21/1, Tatton Asset Management's 2025 and 2026 annual reports and the Bank of England's July 2026 Financial Stability Report. We wrote the questions and answer key; the method and every answer are available on request. Kelriva is not affiliated with these publishers.
One workspace, your choice of where the AI runs
Fast answers from a leading model
Ask in plain English across every document in your workspace. Uses your own Anthropic or OpenAI key. The text of the pages it reads is sent to that provider.
Answers that never leave the machine
An open model and a local search index run on your own computer or server. Questions, documents and answers stay on the hardware you control.
The same fields from every document
Pick the documents, name the fields you need, and get structured results you can download as JSON. Each document is processed on its own, locally.
Trust comes from checking, not from the summary
- Every answer cites its page
- Answers name the file and the page they came from, so a reviewer can check the source in seconds instead of trusting the summary.
- Separate, private workspaces
- Each account has its own files, search index and settings. API keys are encrypted at rest and never shown back in full.
- Read-only by design
- The assistant can read your documents but cannot change, add or delete anything, so instructions hidden inside a document cannot alter your files.
Frequently asked questions
Where are my documents stored?
In a private workspace on the machine that runs Kelriva Cite: your own computer, your company server, or for the demo, Kelriva's server in London. In private chat and field extraction nothing is sent to an outside AI service. In cloud chat, the text of the pages the assistant reads goes to the AI provider whose key you use.
Can I put client documents into the demo?
No. The demo is for trying the product with public or sample documents only. For client or confidential material we install Kelriva Cite on hardware you control and set it up for private mode.
How do I check an answer is right?
Every answer names the document and page it relied on, and you can open the PDF from the workspace to read that page yourself. If the documents do not contain the answer, it should say so rather than guess.
What hardware does private mode need?
A modern computer or server with enough memory for the local models. We size this with you before an install, based on how many documents and users you have.
How do we get it for our firm?
Email info@kelriva.ai. We start with a short call about your documents and where they must stay, then install and configure Kelriva Cite on your own hardware.