AI SDK SSE heartbeat vs Sume jobs_wait 55s slices

The AI SDK heartbeat keeps your browser-facing UI message stream open while it is idle. It does not change how long one Sume jobs_wait call may block. Over remote MCP each call holds at most 55 seconds, so a long video needs several calls with the same job ids.

The release note is from the Vercel AI releases page (ai@7.0.123: "keep idle UI message streams open with optional SSE heartbeats"). Job behavior is from Sume's Jobs and results, read 2026-09-30. The release note gives no option names, so see the package docs for how to enable it.

Which connection does each one affect?

What do I do when a slice expires?

Retry jobs_wait with the same ids. The docs say never to resubmit the paid create. Resubmitting is the costly mistake, because the first job is still running.

What if I wait on several jobs?

Use job_ids (1 to 20) in one batch wait. With wait_for: "any", the response still reports every id, and the remaining jobs continue and still bill. Related: closing the MCP client in onEnd.

Can a heartbeat make a wait longer?

Nothing in either source says so. The docs add that an HTTP request held for 600 seconds dies at the edge with a 502 or Transport send error before it can answer, which is why the server caps slices.