Zed prevent_idle_sleep for agent threads and Sume jobs_wait
Zed 1.21 keeps your machine awake while agent threads run. A sleeping laptop is not a cancelled Sume job: jobs keep billing, so resume with jobs_wait.

Zed's agent.prevent_idle_sleep setting, on by default, stops idle system sleep while agent threads are running, which keeps a long Sume render from being interrupted by a sleeping laptop. Even so, a lost connection does not cancel a job: a Sume job that was already created keeps running and billing, and you read it with jobs_wait or by polling.
What Zed shipped
| Release | Item |
|---|---|
| 1.21 | agent.prevent_idle_sleep on by default, prevents idle system sleep while agent threads run |
| 1.22 | Subagent compaction and a spawn_agent model parameter |
Why it matters for video work
A generation can take longer than a thread's attention span. Without the setting, a laptop could sleep while the model waits on jobs_wait, drop the connection, and leave the thread in an uncertain state. With it, the thread continues. It only prevents idle sleep; closing the lid or a network change still breaks the call.
Connecting Zed to Sume
Zed supports remote MCP servers. Add https://mcp.sume.com/mcp with a Sume API key header, or use OAuth with mcp:read and opt-in mcp:write. Write and paid tools require idempotency_key. Under read-only, they return insufficient_scope.
jobs_waitholds at most 55 seconds per call; the default is 50.- On
wait_slice_expiredcall again with the same ids, never resubmit the create. - Batch up to 20
job_idsand useinclude_results: true.
Subagents and cost
With spawn_agent choosing a model, a subagent may call Sume too. Give each subagent a distinct idempotency_key namespace and a max_spend_usd, so parallel work cannot overspend. Remember that your plan sets concurrency: Free 1, Pro 4, Startup 8, Scale 20.
Sources
Related posts
More in Developers
- 20-minute Reel narration vs Sume's 20,000-character TTS limit
Instagram allows Reels up to 20 minutes. Sume TTS accepts 1 to 20,000 characters per request. A rough estimate of how much narration one request covers.
- 40-clip montage Reel: how Timeline chunks a render past 12 slots
Timeline 1.0 accepts up to 200 slots and chunks automatically past 12; forcing single is refused. What that means for a fast-cut Reel and its $0.10 per minute.
- Retry by error code, not by 5xx: Sume's 502 is your input
Sume's 502 attachment_fetch_failed means fix your URL, not retry. A retry policy keyed on status and code, with a table of which create errors repeat safely.
- 9:16 story image from Sume, then resize to 1080x1920 with Pillow
The v1 image API does not serve explicit pixel sizes, so ask for 9:16, then cover-crop to 1080x1920 in Pillow. Includes the script, models and prices.
Written by Sume