When the session is actively processing user messages, system events (errors, warnings, restart notifications) queue behind them and can arrive minutes late. In one case, a gateway restart error from 16:12 didn't surface until 16:17 because the session was continuously busy with user chat.
This makes error notifications unreliable during active use, which is exactly when they matter most.
Possible approaches:
- Priority lane so system events skip ahead of queued user messages
- Separate delivery channel for errors that bypasses the session queue entirely
- At minimum, visible timestamps on system events so the delay is obvious
Current behavior: system events enter the same FIFO queue as user messages and get indefinitely deferred during active conversation.
When the session is actively processing user messages, system events (errors, warnings, restart notifications) queue behind them and can arrive minutes late. In one case, a gateway restart error from 16:12 didn't surface until 16:17 because the session was continuously busy with user chat.
This makes error notifications unreliable during active use, which is exactly when they matter most.
Possible approaches:
Current behavior: system events enter the same FIFO queue as user messages and get indefinitely deferred during active conversation.