It leaves the building
Every call to an outside AI service takes your data somewhere you cannot see.
Precepta · In development
Precepta is the AI governance product we are building. It is designed to sit in front of every model a company uses, so each request is checked against the company's rules, stripped of anything sensitive, sent to the right model and written down. It is made to run on your own servers, so nothing leaves.
The problem
Every call to an outside AI service takes your data somewhere you cannot see.
Each team writes its own limits into prompts, which the AI can be talked around.
What was asked, what the rules said, what the AI saw: none of it can be pieced back together.
How it is designed
Every request takes the same checked path, before and after it reaches a model. There is no way round it, so a rule set once holds for every team and every tool.
Firewall → Sensitivity → Policy → Smart Route → Cache → Inference → Audit
Which models each team may use, and how much it may spend. The limits sit outside the model, so it cannot talk its way around them.
Personal data is removed before a model sees the request, and the answer is checked for leaks on the way back.
Each request, the decision taken on it and the answer are written down and linked, so you can say what happened months later.
No public route, no telemetry, no phone-home, and checks that the way out really is closed.
Ask for auto and the request goes to the best model available for it, weighing quality, cost and speed.
It is designed to speak the OpenAI API, so existing code would point at it by changing an address. Nothing else.
The cost side
Anything that sees every request can do more than check it. Being in that path is what will let Precepta bring the bill down too.
Not every question needs your most expensive model. Auto is designed to send each one to the cheapest model that can answer it well.
A request answered before would be served from cache, so you do not pay to run it again.
Every key is to carry its own scopes and spending caps. Nobody finds out about a runaway job from the invoice.
Where it runs
Precepta governs the boundary, so it has to sit inside it: in your cloud account, your data centre, or a network with no outside connection at all. Nothing reaches us.
Who it is for
The people who have to answer for what the AI touched, and who already own the servers it should run on. It is designed to govern any agents a company runs, Intent Studio's included, without needing them.
Email us and we will write when the first version of Precepta is ready. If your company has a governance problem it should solve, tell us about it: it shapes what we build first.