Genie One – Unable to Extract Usage Details

devyani_k
Databricks Partner

2 weeks ago

We are looking to create a Genie usage report covering natural language querying activity across three channels — the Genie One web app, Genie One Teams app, and Genie One mobile app. The report is intended to surface insights around user activity, question types, query patterns, feedback, and session behaviour.

What we are currently able to extract via system tables (across all channels)

- User email
- Number of questions asked
- Source breakdown by channel (Web App / Teams / Mobile) with question counts
- First and last active date
- Feedback count

What we are unable to extract

For questions asked across all three channels (web app, Teams, and mobile), we are currently unable to retrieve:

- Actual question text asked by the user
- SQL queries executed in response to the question
- Tables and assets queried
- Feedback type (thumbs up / thumbs down) and associated feedback text where provided

Questions

1. Is there a system table that currently stores this information, or a mapping that would allow us to extract it?
2. If not, is there an alternative approach recommended by Databricks to access this data?

0 Kudos

Lu_Wang_ENB_DBX
Databricks Employee
Databricks Employee

2 weeks ago

1. Is there a system table for this today?

Answer: not as a single supported system table. Databricks system tables currently give you usage/billing, audit events, and executed SQL history, but not a complete per-message record containing the original natural-language prompt, generated SQL, queried assets, and detailed feedback text across Genie One web/Teams/mobile.

What you get today:

  • Paid Genie usage by channel/surface from system.billing.usage, including usage_metadata.genie.surface; only paid usage appears there, not free-allowance usage.
  • Executed SQL from system.query.history.statement_text, and you can identify Genie-originated executions via query_source.genie_space_id.
  • Feedback rating from audit logs via system.access.audit on updateConversationMessageFeedback, which includes feedback_rating.
  • Feedback comments via audit events like createConversationMessageComment, but audit logs expose IDs/events, not the actual prompt or SQL text.

What you cannot reliably get from system tables alone:

  • Original question text asked by the user.
  • A documented mapping from prompt → generated SQL → assets queried across native Genie One channels.
  • Feedback text in system tables.

2. Recommendation

Use the Genie Conversation API for any channel/integration you control, and log the payloads yourself. The API returns the message content (prompt), generated query, query results, conversation/message IDs, and supports listing conversations/messages/comments.

Practical recommendation:

  • Keep using system tables for adoption/cost/channel rollups.
  • Use Conversation API + your own logging if you need prompt text, generated SQL, assets, and feedback/comment detail at report level.
  • For native Genie UI review, use the Monitor experience; it shows saved questions, responses, and feedback, but it is not a system-table export.

Summary: No, not via a documented system table alone today; use Conversation API + custom logging for full-fidelity reporting.

0 Kudos