Skip to main content
workers

​
What Are Workers?

Workers are long-lived processes that execute your agent code. Each agent has a pool of workers.

​
How Workers Stay Warm

Unlike serverless (where containers die after each request), Orpheus workers:
  1. Start when you deploy
  2. Handle requests one at a time
  3. Stay alive between requests
  4. Only die when idle too long
This means no cold starts for subsequent requests.

​
Worker Pool

Each agent has a pool with:
  • min_workers: Always keep this many ready
  • max_workers: Never exceed this many
scaling:
  min_workers: 1   # Always 1 warm
  max_workers: 10  # Scale up to 10

​
Worker Lifecycle

Deploy → Create Pool → Spawn min_workers
                              ↓
Request → Queue → Worker picks up → Execute → Return to pool
                              ↓
Idle timeout → Remove worker (down to min_workers)

​
Check Worker Status

orpheus stats my-agent
Shows:
  • current: How many workers running
  • healthy: How many ready for requests
  • queue.depth: Requests waiting

​
When Workers Scale

  • Scale up: Queue building up (more requests than workers can handle)
  • Scale down: Workers idle (no requests for a while)
See Autoscaling for details.