Run Workbench Ai

Run every configured AI pass over the table. SPENDS CREDITS.

OMITTING row_ids RUNS EVERY ROW, and the body is optional - so a bare POST with no body at all is "enrich this entire table". On a 10,000-row table with three configured columns that is 30,000 model calls.

Set approved_max_credits. The run refuses to exceed it, and a credit preflight happens before any work starts, so an over-budget request costs nothing rather than costing until the balance runs out.

Returns 202 and a handle: the work happens in the background. The group's skip_existing_values and skip_recently_enriched settings still apply, so re-running is usually far cheaper than the first pass.