Skip to main content
A worksheet answers a one-off question: why customers contacted you last week, which deals mention a competitor, how your AI agent handles refunds. To judge every new record as it arrives, on an ongoing basis, use an AI table instead.

​
How it works

  1. Find the data. Use the Query data operations to pick a source table and its columns.
  2. Save a slice. create_worksheet copies up to 10,000 filtered or sampled rows (and linked conversations) into a worksheet. The source table is unchanged.
  3. Add AI columns. enrich_worksheet adds columns that classify, score or extract from each row’s conversation. It returns a job id right away.
  4. Wait for results. get_enrich_status reports progress until it returns done: true; done: false means still running, not failed. The new columns can then be read, filtered and aggregated like any other.
  5. Export. export_worksheet writes chosen columns to a CSV file and returns a download link.

​
Example

Which of last month’s Zendesk tickets were about billing, and how angry was the customer?
  1. list_data_sources with provider: "zendesk" → the ticket table id.
  2. create_worksheet with a where on the created date and sampleRows: 500.
  3. enrich_worksheet with two columns: topic as an enum, frustration as an integer from 1 to 5.
  4. get_enrich_status until done: true.
  5. aggregate_table on the worksheet, grouped by topic, averaging frustration.

​
Good to know

  • Columns requested in one enrich_worksheet call are filled in a single LLM call per row. Request related columns together.
  • Only rows already on the worksheet are enriched.
  • export_worksheet exports every row matching where, not what’s filtered on screen.
  • A CSV uploaded with the Rippit CLI lands as a worksheet too.

​
Operations

See the operation pages in this section for every parameter.