How SmartBot AI agents work
Understand how SmartBot's AI agent orchestration layer classifies guest intents, selects specialized agents, and manages multi-turn conversations.
Audience: Administrators and technical users who want to understand SmartBot's AI agent architecture.
Before you begin
- No configuration is required to use this article. It explains how SmartBot's AI agents function so you can better understand conversation behavior and configure prompts effectively.
How AI agent orchestration works
Intent classification
When a guest message arrives, SmartBot's intent classifier determines the guest's goal. The classifier considers:
- The guest's current message.
- The current active agent.
- The current conversation status.
- The previous SmartBot response.
Based on this analysis, the classifier selects the most appropriate specialized agent to handle the request.
Specialized agents
SmartBot routes conversations to one of six specialized AI agents based on the classified intent:
| Agent | Purpose |
|---|---|
| Booking Agent | Guides guests through the full appointment booking flow. |
| Cancellation Agent | Processes appointment cancellations. |
| Reschedule Agent | Handles rescheduling of existing appointments. |
| Manage Appointments Agent | Lets guests view and take action on upcoming appointments. |
| View Services Agent | Provides information about available services, pricing, and durations. |
| Support Agent | Answers guest questions using the FAQ knowledge base and handles general inquiries. |
Conversation threads per agent
Each agent maintains its own conversation thread. When a guest switches from one task to another (for example, from asking a general question to booking an appointment), SmartBot transitions to the appropriate agent. Threads are reused when returning to the same agent within a session.
Context passed to agents
SmartBot provides each agent with the context needed to personalize responses:
- Session context β Session ID, channel, phone information, and guest profile data.
- Business context β Bot name, organization name, booking rules, timezone, and currency.
- Conversation history β The last three messages of the conversation for continuity.
Handling FAQ questions mid-flow
If a guest asks an unrelated question while a booking, cancellation, or rescheduling flow is active, the Support Agent answers the question using the FAQ knowledge base. Once the question is answered, control returns to the previously active agent so the original flow can continue.
Greeting caching
SmartBot caches responses to greeting intents for a short period to reduce processing time for repeated greetings within a session.
Fallback behavior
When a primary AI service is unavailable, SmartBot automatically switches to a fallback AI service to maintain availability. Retry logic applies for transient errors, with a maximum of three attempts before falling back.