What Sessions Do
Sessions route requests with the same ID to the same worker.Why Use Sessions
Workers keep state in memory. Same worker = same state.- Loaded models stay loaded
- Conversation context persists
- Caches remain warm
How It Works
- Request arrives with
session_id - Orpheus checks which worker last handled that session
- Routes to that worker (if available)
- Falls back to any worker if busy
Best-Effort
Sessions are best-effort, not guaranteed:- If preferred worker is busy → another handles it
- If worker dies → new worker, fresh state
When to Use
| Need | Solution |
|---|---|
| Fast in-memory state | Sessions |
| Durable state | Workspace |
| Both | Sessions + Workspace |

