Most AI tools answer first and explain later — if you ask. Redijo does the opposite: the answer only exists if the proof exists. The agent doesn’t start with the polished sentence; it starts with the source.
The rule is proof-first
Before writing any number, the agent locates the data in a file you uploaded and keeps where it came from — file · sheet · range. The number and its origin travel together through the whole report. Click the value, see the source. That’s auditable by construction, not by trust.
The agent also separates what was observed in your files from what was calculated from them. A margin isn’t raw data — it’s a computation. The report shows both: the calculated value and the sources that went into it.
And when the data isn’t there?
Here’s the detail that changes everything: when a file or a row is missing, the agent doesn’t fill the gap with a guess. It flags what was absent and continues with what exists.
- If June’s statement hasn’t arrived, the month’s cash position is marked pending — not estimated.
- If a sheet was renamed, the agent warns you before calculating, instead of summing the wrong column.
A number you can’t trace isn’t a number you can trust. So the agent would rather say “this is missing” than hand you a total that looks right and isn’t. Each fragment’s permission also holds at synthesis time: what you can’t see never enters the answer — not even hidden inside a total.