Data posture
Agent: the posture agent (ai_posture), behind PostureService.
What it does
It reads your connected tables and documents, groups them into business domains, and describes what each domain’s data is missing. Its jobs:
- Map the estate. Groups tables that share keys, entities or subject into domains, and names how each domain relates to the others, including links of business meaning where no key is shared.
- Describe each domain. Its reading of a domain is shown as the model’s. Grades, readiness percentages and ratings it writes are not shown: only what the profiler measures is (see Posture & maturity).
- Name the incidents. Writes up material data-maturity incidents with severity, impact, root cause, blast radius and a recommended fix.
What it reads
Table schemas, sample rows and extracted documents from the connections you configure. Exploring a source never samples customer rows; reading rows is a separate, bounded call (see Data sources).
What it produces
A posture snapshot: domains, their relationships, a maturity scorecard and incidents, shown on Why. Results the agent wrote are marked as model-generated. See Posture & maturity.
Human oversight
- The data owner decides on every recommendation.
- The agent never changes schemas, masking or access.
Model
The deployment’s general reasoning tier.
What FACE measures about a domain comes from a deterministic path that uses no model; Posture & maturity says which result came from where.
Where it runs and data handling
On the deployment’s own model plane: Runink’s shared machines on a Trial, your own cloud on Cloud and your own servers on On-premises. No outside AI service is called on any plan. See Facts shared by every card.
Guardrails
- Builds the domain map from your schemas and samples. It doesn’t invent tables or domains.
- Refuses to inflate readiness scores or data-quality grades without evidence.
- Refuses to suggest or test dangerously open security settings, or to alter audit logs.
- Doesn’t penalise vendors or regions on assumptions rather than measured evidence.
- Stays on posture: fulfilment and claims belong to other agents.
How to read its results
- A grade reflects the structure of the data it could read, not how the data is used.
- A failed scan is an error, not a grade of F. A domain that couldn’t be read is not assessed, which is different from a pass.
- Business-meaning links between domains are the model’s inference. Check them before relying on them.