Prompts and conversions have separate list/get endpoints. List endpoints return their server page and, where supported,
next_cursor. Pass the cursor with the same filters for the next page. Date filters use RFC3339 strings. Metadata filters use query parameters such as metadata[customer]=acme.
Limits and cancellation
limits controls server work and cost thresholds. Dollar amounts in responses are exact decimal strings. Limits are soft: an active model response, tool, or export can finish beyond a threshold. A client timeout is a separate waiting budget and does not cap billing or issue a refund.
Closing your connection or aborting a client request does not cancel the server job. Use POST /v1/prompt/{id}/cancel or POST /v1/convert/{id}/cancel explicitly, then inspect the returned status.
Eligible paused prompts can resume through POST /v1/prompt/{id}/resume. Supply both max_platform_cost_usd and max_inference_cost_usd as new absolute limits, not increments. The resume body does not accept max_seconds.
TypeScript: wait on an existing request
wait returns paused/failed/cancelled receipts for inspection. run, edit, and convert require success and throw ReviseJobError for those outcomes. Successful partial prompts remain successful; inspect incomplete and stop_reason.
The edit/convert helper deadline covers source resolution, upload, submission, and polling. A returned Source downloads later with its own signal. Pass the same external signal to both operations when you need one end-to-end deadline.