Why we built this
People in finance have good reason to distrust AI answers. We started from that and built something that has to earn the trust.
A wrong number is worse than no number
So when your documents don't say, the answer is "not found". Figures in an answer are checked against their source before you see them, and ones that can't be traced are dropped.
Access is enforced by the database
Which department's documents you can reach is decided where the data is stored, not by which buttons the screen shows. It's the part we test hardest.
Your data stays with you
The model runs on your servers by default. Sending anything to an outside service is something an administrator has to switch on deliberately.
We score ourselves
We keep a couple of hundred test questions with known answers and run them before changes ship. Admins can run the same check on their own installation.
Try it on your own documents.
Bring a few reports and a question you've had trouble answering. We'll work through it with your files, not ours.