Why it matters
A node producing readings continuously would generate an unbounded bill if every reading triggered a model call. Gating is what makes the AI layer affordable - and by making it occasional, it also makes it non-critical.
Systems & architecture
AI gating
AI gating is the practice of invoking an AI service only when defined conditions are met, rather than on every input, to control cost and bound the system's dependence on it.
How it works
Gates are typically severity - only analyse what already matters - a cooldown per device to prevent repeat calls on a persistent condition, and an availability check with a defined fallback.
Example
A node sitting just above a warning threshold for an hour produces one analysis, not one per reading.
In Alfa HazardX
Threshold breaches trigger analysis, subject to a per-device cooldown of sixty seconds. With no model key configured, or the service unreachable, the system falls back to threshold-only output.
Read further
Related terms
- Generative AI in safety systemsGenerative AI in a safety system means using a language model to describe, explain or recommend - as an advisory layer over decisions made elsewhere, not as the mechanism that makes them.
- Deterministic analysisDeterministic analysis produces the same output for the same input every time, using explicit rules or arithmetic rather than a learned model.
- Fail-safeA fail-safe design ensures that when a component fails, the system moves to a safe state rather than an unsafe or unknown one.