Orchestration
The Orchestration panel lets you define rules for agent behavior, escalation paths, and coordination between agents.
What Orchestration Controls
| Rule Type | Purpose |
|---|---|
| Escalation rules | When to escalate to other agents or humans |
| Handoff conditions | Automatic routing based on intent, sentiment, or custom conditions |
| Approval requirements | Which actions need human sign-off before executing |
| Workflow triggers | Multi-agent coordination patterns |
Rule Types
Escalation Rules
Define when an agent should escalate to a supervisor or human. Common triggers include: too many failed attempts, specific intents (billing disputes, legal questions, security incidents), or explicit user request for a human.
Handoff Rules
Automatic routing based on detected conditions. For example, technical support questions go to the tech-support agent, billing inquiries go to the billing agent, and negative sentiment conversations go to the retention agent.
Approval Requirements
For sensitive actions (issuing refunds, modifying accounts, accessing sensitive data), you can require human approval before the agent proceeds. Specify which actions need approval and who can approve them.
Configuring Orchestration Rules
Via the Dashboard
- Open Studio > Orchestration tab
- Select the rule type you want to add (escalation, handoff, or approval)
- Define the conditions that trigger the rule
- Set the target (another agent, a human, or an approval workflow)
- Save and test with sample scenarios
Escalation Configuration
| Setting | Description |
|---|---|
| Max Attempts | Number of failed attempts before escalating |
| Intent Triggers | Specific intents that trigger immediate escalation |
| Target Agent | Which agent receives the escalation |
| Escalate to Human | Whether to route to a human instead of another agent |
Handoff Configuration
| Setting | Description |
|---|---|
| Condition | Expression to evaluate (intent, sentiment, custom data) |
| Target Agent | Which agent receives the handoff |
| Priority | Order in which conditions are evaluated |
Approval Configuration
| Setting | Description |
|---|---|
| Actions | Which actions require approval |
| Approvers | List of users or roles who can approve |
| Timeout | How long to wait before the approval request expires |
Best Practices
- Clear conditions -- make escalation triggers explicit and testable
- Fallback paths -- always have a fallback (usually a human)
- Test thoroughly -- verify rules work as expected with edge cases
- Monitor -- track escalation rates to tune rules over time
Next: Learn about Publishing your templates.