Ask Aria: what the assistant can and can't do

Aria is Veritome's built-in assistant. It knows the EU AI Act and can see the context of the system or page you're on, so you can ask plain-language questions instead of hunting through 113 articles and 13 annexes. Aria runs on Mistral Small, hosted in the EU (Paris / Nuremberg) — your data is not used for training, and no customer compliance data leaves the EU.

Where you'll find Aria

• The full Ask Aria chat page in the sidebar.
• The launcher drawer — open Aria from anywhere; it carries the context of the page you're on.
• Suggestion cards inside the classification wizard, where Aria proposes the risk tier and role with its reasoning.
• Aria's recommendations on the dashboard, and Agent mode, where you give Aria a goal and approve what it proposes.

What Aria is good for

• "Does Annex III cover a CV-screening tool?" — interpreting the Act.
• "What does Article 13 require me to give my deployers?" — explaining an obligation in your situation.
• "Draft the human-oversight section for this system." — a first draft you then review and edit.
• "Why was this classified as high-risk?" — explaining the engine's reasoning.

Asking a question

Open Ask Aria (or the launcher drawer), type your question or click a Try asking suggestion. To ground the answer in your own material, click the attach icon and upload a document (PDF, DOCX, TXT, MD, CSV) — Aria reads it and answers against it. Press Enter; Aria streams its reply with clickable references. Use 👍 / 👎 to rate an answer, Copy to lift it into your work, and the left rail to resume a recent conversation.

Controlling what Aria sees

From Aria's settings you control the context sent with your questions: Read obligation status and Read risk classifications are on by default; Read team member names and Read document titles are off by default; Anonymise system names replaces real names with generic IDs; Save conversation history is off by default. You can also toggle the knowledge sources Aria reasons over (GDPR, ISO/IEC 42001 and others); the EU AI Act itself is always on.

The accept / adjust guardrail

Aria is a drafting and research aid, not a legal sign-off. Every suggestion — a proposed risk tier, a drafted control, a classification, an agent action — is a proposal for a human to confirm. In the classification wizard, Aria's proposed tier and role always land on an explicit accept / adjust step before they take effect. In Agent mode, proposed actions queue for your approval. Regulator-facing actions (incident notifications, EU database registration, dossier submissions) are never automated — Aria will not even propose them. Veritome records who confirmed what, so the human decision — not Aria's suggestion — is the one in your audit trail.