Jobs

A job is a unit of work — a prompt, optional context, and configuration — that gets queued, assigned to a worker, processed, and stored.

Job states

State	Description
pending	Queued, waiting for a worker
running	Assigned to a worker, in progress
done	Finished successfully, result available
failed	Errored during processing or aborted by operator

Dispatching a job

From the saddle (VSCode extension): Type your prompt in the text area and hit “send it.” The router picks the best available worker based on your effort tier and mode settings.

From the dashboard: Use the chat panel at app.modelreins.com.

Via API:

curl -X POST https://app.modelreins.com/dispatch \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Review this code for security issues"}'

Job priority

The saddle’s effort tier controls job priority:

Tier	Routing behavior
trivial	Cheapest available, local preferred
quick	Fast + cheap
standard	Balanced (default)
deep	Best quality, cloud preferred
critical	Best available, no cost constraint

Retrieving results

Results stream back into the saddle or dashboard automatically. Via API:

# Get job status + output
curl https://app.modelreins.com/jobs/JOB_ID \
  -H "Authorization: Bearer YOUR_API_KEY"

Fan-out dispatch

Select two or more workers in the saddle’s target picker to dispatch the same prompt to all of them at once. Each answer streams into its own card, side by side. A merged canonical answer appears when all workers finish.

Aborting jobs

Hit the red kill button in the saddle to abort all in-flight jobs in the current thread. Or hit KILL ALL in the dashboard to abort everything across your tenant.

Automatic retries

If a worker crashes mid-job, the coordinator requeues the job after the configured timeout. See Troubleshooting for stuck job handling.