Conversational analytics
Is it safe for our data?
Mark Your Data Assistant reads your warehouse with a read-only account and checks every query before it runs. You choose the AI providers, and you configure what they see and what the assistant stores, so your company can make its own assessment.
How it works out of the box
Read-only, with checks. The assistant uses a read-only account. A SQL guard allows one SELECT statement on approved tables, with limits on the data it fetches.
You choose the AI partners. You configure the assistant with providers you contract directly, and their terms set retention, training use and processing location.
You configure what the model sees. With Structure only, the model gets column names, column types and the number of rows. With With results, the model that writes the answer also reads the first rows.
You configure what is stored. Records can hold metadata only, text with names and other values from your data masked, or everything as is.
You can hide sensitive columns. Mark a column in dbt, and the SQL guard refuses it or keeps its values out of results.
What it means for you
- Manager: Your company chooses the partners, and we go through the technical setup with your legal and security teams.
- Analytics engineer: You control access through the warehouse, and mark sensitive columns in dbt.
- Business user: The assistant tells you that it's an AI, and shows where its numbers come from.
Questions and answers
Can the assistant change or delete data in our warehouse?
No. Mark Your Data Assistant queries your warehouse through a read-only account, so it has no permission to change or delete data. As a second line of defence, every query must first pass a SQL guard, which allows one SELECT statement on approved tables. The AI never runs SQL itself.
Does our data leave our cloud?
Yes, your questions go to the AI providers you choose. You configure Mark Your Data Assistant with providers you contract directly, such as Gemini on Vertex AI, OpenAI or Mistral, and a decision model such as TypeSafe Jev or an LLM for routing. Their terms set retention, training use and processing location.
Do users only see the data they're allowed to see?
Not per user yet. Mark Your Data Assistant queries through one read-only account today, so every user can reach what that account can read, and you decide what that is. Signing users in through the platform's own authentication, so its access rules apply per user, is on the roadmap.
What does Mark Your Data receive from our installation?
Mark Your Data will collect monthly usage totals from your installation, such as active users, questions and tokens, to support your installation and keep improving the product. These totals contain no query results, question text, SQL, answers or user IDs. Anything more sensitive will be collected only if you explicitly want that, and pseudonymised.
How do we explain the privacy setup to our legal and security teams?
Mark Your Data goes through the technical setup with your legal, security and compliance teams, for your installation. That conversation covers three decisions: what the AI services see, what the records keep, and what goes to Mark Your Data, along with the providers you choose.