Resume a Gemini CLI session after a Sume render: keep the job ids
Gemini CLI 0.64 preview stops deleting resumed session history. How to pick a Sume video job back up after a restart with jobs_list and jobs_wait.

If a Gemini CLI session ends while a Sume render is running, resume the session and ask for the job by id, not by prompt. The 0.64.0 preview release notes dated October 6, 2026 list a session management fix that prevents deletion of resumed session history, so the transcript that holds your job id is more likely to survive a resume. The render itself never lived in the session: Sume runs it on its side.
Write the job id down anyway. A transcript is a convenience; a note is a safeguard.
What lives where
The split below is why resume works at all.
| State | Held by | After a restart |
|---|---|---|
| Chat transcript | Gemini CLI session | Available if you resume the session; 0.64 preview fixes a history deletion problem |
| Render in progress | Sume job queue | Keeps running; read it with jobs_list or jobs_wait |
| Job id | Tool result in the transcript | Copy it out before closing the session |
| Auth | Header or OAuth session | An OAuth token lasts 3600 seconds with no refresh token; sign in again |
Resume steps
Resume the CLI session. Ask the agent to call jobs_list and name the job that matches the clip you started. Then call jobs_wait with that job id. Sume's docs give jobs_wait a bounded wait of up to 55 seconds per call, so for a long render, call it again until the status finishes. When it completes, read the result with jobs_result and save the output file you need to your own storage.
- Do not resubmit a render because the transcript looks quiet; you would pay twice.
- If you must resend a submit, reuse the same
idempotency_keyso the call is a duplicate rather than a new job. - Check
balance_getafter a long run.
A note format that helps
Keep a short line per job in a file next to the project: the job id, the tool you called, the idempotency key you used and the time. Four fields are enough. If the session is lost, that line is all you need to call jobs_wait again, and the idempotency key lets you recognize a duplicate if you decide to resubmit.
Do not paste your API key into the note. The job id and key are not secrets in the same sense, but the file may end up in version control, so keep it out of the repository or ignore it.
A caution about preview builds
The release is a preview, and the notes describe a fix, not a guarantee. Do not make your only record of a paid job a transcript that lives in a preview build. The note file and Sume's own job list are the sources you can count on; the resumed history is a bonus.
What Sume does not do
Sume does not attach a job to a CLI session or restore your chat. Its docs do not describe any cancel-on-disconnect behavior; to stop a job you call jobs_cancel, a write tool that needs its own idempotency_key.
Sources
Related posts
More in Agents
- After generate_image on sume/auto, why can't the agent name the model?
Sume never discloses which family ran for sume/auto: job.model stays sume/auto and the model list omits it. To name a model, pin an id from image-models_list.
- Agent leaves model blank on generate_video: 720p 8 s costs $1.00
Omit payload.model and generate_video routes to sume/auto: 3 to 10 s, 720p and 8 s by default. At 720p the price runs from $0.38 for 3 s to $1.25 for 10 s.
- Haiku 5.5 effort for an agent that polls Sume jobs
Claude Haiku 5.5 defaults to medium effort. What that means for a loop that calls Sume jobs_wait, and when to move to low or high.
- Use a low-effort Haiku 5.5 subagent for Sume dry_run previews
Anthropic recommends low effort for subagents, and warns it can skip checks. Where a Haiku 5.5 subagent fits in front of a paid Sume call.
Written by Sume