Reliability

The Dispatch system includes built-in reliability features to ensure agents complete their work.

Why Reliability Matters

Agents can fail for many reasons:

• API rate limits
• Network issues
• Model errors
• Timeouts

Without reliability, failed tasks are lost.

Automatic Retries

Failed tasks are automatically retried with exponential backoff:

1. First retry - 1 second delay
2. Second retry - 2 seconds
3. Third retry - 4 seconds
4. And so on...

This prevents overwhelming services while ensuring eventual completion.

State Persistence

Workflow state is persisted across:

• Server restarts
• Deployments
• Failures

A workflow can pause, the server can restart, and it picks up where it left off.

Idempotent Execution

Steps are designed to be safely re-run:

• Same input → Same output
• No duplicate side effects
• Safe to retry

Error Handling

When retries are exhausted:

• Fallback steps can execute
• Notifications are sent
• Tasks are marked failed
• Humans can intervene

Monitoring

Track execution health:

• Success/failure rates
• Average execution time
• Retry frequency
• Error patterns