TellParrotEvery AI governance framework starts with the model. Register it, classify it, name an owner, assess its risk, document its purpose. All of that is right. None of it answers the question that decides whether the model is safe to use: what is it reading?
A model with a complete file, reading data whose retention has lapsed, whose consent is unclear and whose duplicates were merged by a score, produces confident, well-documented mistakes.
The register, done properly
TellParrot keeps a complete register: AI initiatives, projects, proposals, models and agents, each with an owner and a system classification. Risks and controls map to the EU AI Act and the NIST AI Risk Management Framework. Impact assessments, an incident register, public disclosures, and contestability and appeals workflows sit alongside, so the portfolio-level posture is a set of documents you could hand to an auditor rather than a set of intentions.
The score that connects the two halves
The part that is distinctive is a single number on every model: a weighted data-compliance score derived from the governance state of the data the model reads. If the sources feeding a model are poorly governed, the model's score says so, and it changes as the data does. A data-retention compliance view shows the same relationship from the other side: which models are reading data that should already be gone.
This is only possible because the AI register and the data governance pillar share one knowledge graph. In separate products, the score is a spreadsheet someone updates before the review.
Agents under approval
TellParrot uses AI inside the platform, and it holds itself to the same rule. The onboarding assistant, the MDM agents, the governance advisor and the copilot all work by proposing, and a person approves. The ledger records who approved what. Agents never act silently, and the advisor cites its sources from the tenant's own data.
AI provider keys are yours, scoped per tenant, whether that is Azure OpenAI, OpenAI, Claude or Gemini. No tenant data leaves through a shared key.
The layers under the model
The AI readiness view lays out the thirteen layers underneath a model, from source systems up to tribal knowledge, and names which pillar owns each one. It is the honest version of the "AI ready" slide: the version that shows how much of the stack sits below the model, and how much of it is data governance by another name.