DecisionManager

AI Intelligence

Grounded AI across the entire decision lifecycle.

Four AI clusters — authoring, testing, operations, DevOps — all grounded in platform telemetry. No hallucinated evidence. Every AI call is optional, priced before it runs, and joined onto measured data so omissions are visible rather than absent.

Four grounding principles

Measurement before narration

Every AI panel calls the platform's measurement APIs first. The model narrates what the data shows — it does not generate the data.

Refusal is honest

Each panel has explicit refusal conditions: outcome drift AI refuses if drift is unmeasured, hotspots AI refuses if no telemetry exists. These are documented in the UI, not hidden.

Omission is visible

When a model's answer is joined onto a measured list, items the model missed render as gaps — not absences. A blast radius that silently drops a consumer reads as an all-clear, which is worse than no blast radius.

Evidence is separable

In every export (RCA, evidence pack), model prose is fenced between markers. A security reviewer can strip the AI narration — the measured evidence stands on its own.

Four AI clusters

All AI clusters included on every plan

AI panels are opt-in and click-only — no model calls on mount. Token cost is estimated and shown before any call is made. AI provider configuration requires Business or Custom plan.