You added AI to your product. Now what?
You know the AI is going to be wrong sometimes. You’re not sure what to do about that. We’ll tell you.
Here’s what’s actually going on
- Your AI feature shipped. Or it’s about to. Your competitor already has one.
- Sometimes the AI says things that aren’t right. You know this. Your users might not.
- Someone — an investor, a customer, a lawyer — is going to ask how you handle that.
- You don’t have a compliance team. You don’t have a governance framework. You have a TODO.
- That’s fine. Most companies at your stage are in the same spot.
What Ontic does, in plain terms
Ontic sits between your app and your AI model. When the model is about to say something, Ontic checks whether the answer is backed by your actual data. If it is, the answer goes through. If it isn’t, it gets blocked — and the system tells the user what it would need to answer properly.
Everything gets logged: what was asked, what was checked, what went through, what got blocked. That log is your evidence that you have controls in place.
Three environments. Start with the first one.
The Studio
The AI answers like normal, but every output gets labeled as verified or unverified. Everything is logged.
Effort: 1 hour to set up. Free.
When: You’re just getting started. This is your minimum viable governance.
ontic init --template minimalThe Refinery
The AI can only answer if Ontic finds evidence in your data. No evidence, no emission.
Effort: A few days. Requires connecting your data.
When: Your AI talks to customers or generates content people act on.
ontic init --template financial # or medical, legalThe Clean Room
Every step is cryptographically signed. Auditors and courts can trace every decision.
Effort: Weeks. Requires architecture planning.
When: You’re in healthcare, life safety, or serious regulatory territory.
ontic init --template life-safetyIf this is you, here’s what to do
“We have an AI chatbot on our website”
Risk: It'll answer questions wrong. Customers will notice.
Do this: Start with The Studio (labeling). Upgrade to The Refinery when you can connect your knowledge base as a data source. Try the Oracle Builder to find yours.
Effort: 1 hour today. A few days next sprint.
“We generate reports or summaries with AI”
Risk: Wrong numbers. Wrong facts. Wrong recommendations.
Do this: Start with The Refinery. Connect your data source. The AI checks its claims before the report goes out.
Effort: A few days.
“We're in healthcare/fintech/legal and using AI”
Risk: Regulatory exposure. Audit risk. Liability.
Do this: Run the risk profile wizard immediately. It generates a compliance posture report you can show to investors and regulators. Start with The Refinery, plan for The Clean Room.
Effort: 2 minutes for the report. Days-to-weeks for full setup.
“An investor asked about our AI risk posture”
Risk: You don't have an answer yet.
Do this: Run the risk profile wizard. Download the board-ready risk assessment. Send it to them. Then set up the actual controls.
Effort: 2 minutes for the document. An hour to set up basic controls.
“We just use AI for internal drafts and brainstorming”
Risk: Minimal. A human reviews everything.
Do this: You probably don't need this yet. Bookmark us. Come back when the AI starts talking to customers.
Effort: Zero.
Not sure where you fall? Answer two quick questions.
Check Your Risk