A workflow,
not a model picker.
Sume Agent holds the product, script, host, voice, and edit in one thread — and comes back with a finished show, not a clip. When the workflow settles, we lock it into a Format your systems call over one API.
What are we creating today?
Cases we actually ran.
A Korean live-commerce team builds AI show-host broadcasts in the @mobidoo workspace: product URL and Korean script in — stills, host avatar, TTS, and lip-sync out, bound into a broadcast-length timeline MP4. These are their shows.
Same show.Every SKU. One API.
When the workflow settles, we lock it into a Format — a versioned package of hard rules the agent must follow. @mobidoo/live-commerce pins the host and voice, the banner and price-card layouts, per-scene holds, and the sentence map that binds every script line to its scene. From there, a sheet can call it.
curl -X POST \
https://api.sume.com/v1/formats/mobidoo/live-commerce/runs \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Idempotency-Key: sheet-row-45" \
-d @row.json
{ "object": "format.run", "status": "processing",
"status_url": ".../format-runs/arun_.../status",
"result_url": ".../format-runs/arun_.../result" }One sheet window, queued over the API
Twenty rows from a product sheet went into one run queue against the pinned Format — no chat, no operator in the loop. Completed shows came back in about three to five minutes each.
Read the docs20
sheet rows queued
3–5 min
per completed show
v38
format version pinned
Real runs from one production queue window. Illustrative run economics — not a rate card and not a throughput promise.
Enterprise contracts
Need volume pricing, security review, or a custom agreement? Book a call and we'll scope an enterprise contract for your team.