
The Conversational Analytics API lets a team build a data agent that answers natural-language questions about data such as BigQuery tables. For the Professional Agentic Architect exam, access control for these agents comes down to two separate requirements: granting the right dedicated role on the individual data agent, and ensuring each user holds permission to read the underlying data source.
Access to a Conversational Analytics data agent is controlled with dedicated roles, which Google recommends granting on the individual agent:
roles/geminidataanalytics.dataAgentUser): allows a user to chat with, get, and list data agents.roles/geminidataanalytics.dataAgentViewer): allows only get and list operations. A user with this role can see that an agent exists, but cannot ask it anything or send it a question.roles/geminidataanalytics.dataAgentCreator): allows a user to build new agents. It does not include permission to chat with an existing agent.roles/geminidataanalytics.dataAgentOwner): adds permission to update, share, and delete data agents.Each role covers a distinct set of actions, and several permissions do not carry over from one role to another. Data Agent Viewer allows getting and listing agents, yet it does not allow sending questions to an agent. Data Agent Creator covers building new agents, without granting permission to chat with an existing agent.
Data Agent Owner expands access by adding permission to update, share, and delete an agent. When users only need to ask questions of an existing agent without managing it or building new agents, we would lean toward granting Data Agent User on that individual agent.
A data agent role controls what a user can do with the agent itself, not what data the user can read. The Conversational Analytics API queries the data source with each user's own credentials. Because every query uses the user's own credentials, anyone chatting with a data agent also needs permission to read the underlying tables.
For data stored in BigQuery tables, the setup documentation grants that permission with the BigQuery User role. On the Professional Agentic Architect exam, evaluating access for a Conversational Analytics data agent requires verifying both sides: the dedicated role granted on the individual agent and the user's own permission to read the underlying BigQuery tables.
Our Professional Agentic Architect practice exams test Conversational Analytics data agents alongside data grounding and agent access control, with detailed explanations that drill these distinctions.