Solutions / Governed AI
Your AI answers from approved knowledge.
Assistants and agents read only from entries a person signed off. Every answer shows who approved it, when, and from what source.
The problem
The assistant works. It waits.
“The assistant is built, tested, and funded. It has been sitting in governance review for four months.”
Before an AI system is allowed to go live, a review board asks one question first: where did that answer come from, and who stood behind it.
Building it was never the hard part. Your team finished it, tested it, and has the budget.
Then the board asks which approved source an answer came from. The tool can name a file. It cannot name a person or a date.
So the project waits, and every other team in the business solves the same problem its own way.
What it does
Point the assistant here instead.
Your assistant stops reading a pile of files and starts reading approved entries. Every answer comes back with the approver, the date, and the source version.
Use disclosure D-214 rev 6. It supersedes rev 5 for all retail communications and carries the current market-risk language. SOP-4471 v3
An approved answer card showing the question, the answer, the named approver, the approval date, and the source document and version
How it works
What changes on every answer.
Source
It reads approved entries
Your assistant asks Cognatum instead of asking a folder. Only entries a person approved come back.
Refuse
It says when it does not know
No approved answer means no answer. It will not write one that sounds right.
Record
Every answer names a person
Who approved the source, on what date, and which version. That record is what a reviewer asks you for.
Re-check
A changed source flags its answers
Edit or retire a source and everything built on it is queued for review before it is served again.
In practice
Stuck at the board.
What changes about the assistant, and what does not.
An assistant has passed technical review and is in its fourth month at the board.
Your team repoints it at Cognatum. Nothing about the assistant itself changes, only where it reads from.
Now every answer it gives names the person who approved the source, the date, and the version. Your team exports that record and takes it to the board.
The next team that builds one starts from the same pattern instead of from nothing.
Common questions
What teams ask first.
Does this make our AI compliant?
No. Nothing here confers compliance with any rule or standard. What you get is evidence to hand to the people who ask: a named approver, a date, and a source on every answer. The obligations stay yours.
Which assistants does it work with?
Any that can call a tool. It connects over MCP, the open standard AI assistants use to reach outside data, and over a plain API.
What happens when there is no approved answer?
It says so and stops. The question becomes a ranked gap with an owner, rather than a plausible sentence nobody checked.
Do we have to move our documents?
No. Entries are drawn from the systems where the work already happens, and each entry links back to the file it came from.
Knowledge governed. Intelligence everywhere.
See it on your own content, in your own environment.