AI can never out-alarm the rules
If the rule-based verdict is Warning, the AI cannot report Critical. If no limit is configured, the AI cannot raise an alarm at all. This ceiling is applied in code after every AI answer.
AI governance
AI you can govern
NeoAI uses AI to help explain what your data shows, never to decide what is safe. The guardrails are enforced in the platform's own code, not just written in a policy.
The principle
Engineering limits come from rules. Not from AI.
Operating limits, such as the battery temperature at which an asset becomes Critical, come from governed, rule-based configuration. An AI model is never allowed to originate a limit, override one, or pretend a limit exists when none has been configured.
Preview
Where AI sits today
AI-generated operational insights are an early preview. The guardrails below apply to them now, so the preview is safe to put in front of your team.
Guardrails in the platform
Six guardrails, enforced in code
-
Available today -
Available today Too little data? The AI stays silent
With one reading or none, NeoAI does not ask the AI at all. It records that there is not enough evidence, rather than generating a plausible story.
-
Available today Disagreement stays visible
The AI's own view and the rule-based verdict are stored side by side. Neither is silently rewritten, so a reviewer can always see both.
-
Available today Grounded in your own documents
AI insights draw on your own readings and your own reference documents, such as equipment manuals. Document search is limited to your organisation's data.
-
Available today Limit changes need a second pair of eyes
Configuration changes move through draft, validation, preview, approval and activation. The person who wrote a change cannot approve it, and every past version is kept.
-
Preview Read-only when your AI assistant connects
Early preview
An external AI assistant connected to NeoAI can look things up, but never change anything. It sees only what your account may see, and every request is logged.
Four signals
Facts and opinions are never blended
Every measurement carries four separate answers, so nobody mistakes an AI opinion for a rule-based fact.
- HealthNormal, Warning or Critical, decided by rules.
- TrendRising, falling or stable.
- Data qualityOK, insufficient, or no data.
- AI opinionA clearly labelled view beside the facts, never instead of them. Early preview
Roadmap
The AI roadmap stays governed
Forecasting, anomaly detection and lifecycle analytics are planned on the same foundation. They will stay advisory: a lower-confidence result may add context, but may never overrule a higher-confidence one, and nothing in this layer controls equipment. Not available today.