Google opens BigQuery conversation logging for AI agents
Google Cloud has made conversation logging and its Conversational Analytics interface generally available for BigQuery agents. Organizations can opt to store conversation content in their own tables, separately from sanitized operational telemetry. Access restrictions and encryption protect the stored content, while some feedback, evaluation and daily usage controls remain in preview.
Artificial Intelligence··Evening
Conversation content is logged by choice
Google Cloud made conversation logging generally available on November 14 for BigQuery, its cloud data-analysis service. Organizations can choose to store AI agents’ prompts and replies in tables they control. This content is separate from operational telemetry, which removes prompts, answers, reasoning content and database-query text. The company’s announcement also moves the Conversational Analytics application programming interface into general availability, alongside monitoring dashboards in BigQuery Studio and the Looker analytics service.[1]
Privacy checks can skip a log entry
The optional transcripts go into the customer’s agent_events table. Access permissions, customer-managed encryption keys and Google’s VPC Service Controls data boundaries apply to the stored material. For queries with row-level security, access restrictions on columns or dynamic masking, the service can omit the log entry. The chat continues even when saving a log fails. The same framework collects information from Google agents and custom agents built with its Agent Development Kit or the LangGraph development framework.[1]
Some evaluation and quota tools remain in preview
The announcement separates shipped monitoring from features still being tested. Daily input and output token quotas, feedback ratings and evaluation within BigQuery remain in preview. Administrators can already limit the data scanned by an individual BigQuery query. Prism, Google’s open-source agent-testing tool, can evaluate agents using BigQuery tables or Looker data views and send results directly into BigQuery. Custom agents can also expose the structure of tool calls and interactions involving multiple media types.[1]