Agents & AI

How agents work

An AI agent takes your question, plans the steps needed to answer it, and executes them — looking at your schema, glossary, domain context, and available data sources to build and run the right query. It is what turns a plain-language question into a trustworthy, executed analysis.

Agents work with the guardrails your workspace has configured: they only access data you have permission to see, and every query they produce is shown to you before it runs.

Choose a model

You can pick the AI model used for your conversations, either per workspace or per conversation. Different models have different strengths in terms of speed, SQL accuracy, and reasoning on complex questions.

The best choice depends on your workload: for quick, simple questions a fast model is often enough; for complex analytical questions you may want a stronger reasoning model. You can switch at any time and compare results.

Monitoring

Administrators can monitor agent activity across the workspace — the questions being asked, the queries generated and executed, and their success. This gives visibility into how the assistant is used and where accuracy can be improved.

Monitoring data feeds back into the platform: recurring patterns, saved golden SQL, and glossary updates all help the assistant get better over time while keeping usage fully transparent.