Make AI Agent fallback connection retries once: key Sume calls

If a Make AI Agent calls a Sume HTTP tool and its LLM connection fails mid-run, Make now retries the run once on a fallback connection. Whether tools already called are repeated is not stated, so build for a second run: send an Idempotency-Key that comes from your scenario data, never from the model.

Make facts are from its September 30 feature spotlight; Sume facts from Communication modes and Errors and credits, read 2026-10-01.

What did Make add?

The spotlight says that if the primary LLM connection fails, Make automatically retries the run once using the fallback connection, which defaults to the Make AI Provider. You set it under "Show advanced settings" on the AI Agent module. The post does not describe what state, tool calls included, carries over to the retry.

Why would a retry double-bill?

A second run can reach the same decision and call the same Sume tool again. A random or model-written key would differ each time, so each call would look like a new submit. A key built from a stable id means the repeat is recognized: an exact retry returns the original job instead of billing twice.

How do you set it in the HTTP module?

Map the key into the header from the scenario input. This is the request the module should send; the {{1.orderId}} token is Make's mapping from the first module.

curl -X POST https://api.sume.com/v1/video-1.0/generate \
  -H "x-api-key: $SUME_API_KEY" \
  -H "content-type: application/json" \
  -H "idempotency-key: video-{{1.orderId}}-r1" \
  -d '{"prompt":"Slow push-in on a ceramic mug","mode":"async"}'

What should the agent be allowed to decide?

Let it choose the prompt and whether to submit, but not the key. Reuse a key only for the same operation and payload. After the submit, poll GET /v1/jobs/{id}/status or take the signed webhook; for stuck webhook items see Make's queue and redelivery.