​

AI Agent Block

​

How It Works

• The AI Agent block waits for an incoming user message and immediately generates a reply.
• Each time the user sends a new message, the agent runs again (multi-turn loop).
• The agent uses:
  • The system prompt (personality + rules)
  • The connected knowledge base (grounded context for answering)
  • The last ~20 conversation messages (recent chat history)
  • The customer context (contact name/email/phone + saved attributes)
• If an intent route is triggered, the automation exits the agent and continues from the connected route path.

​

Key Components

1. System Prompt Define the agent’s role, tone, and behavioral rules. This prompt applies to every message the agent generates.
   Use Generate with AI to create a strong starting prompt from your instructions, then refine it for your business.
2. Connected Knowledge Base Select a knowledge base for the agent to reference while answering questions.
   • The agent automatically searches the selected knowledge base for each user message.
   • This helps keep answers more accurate and consistent with your business content.
3. Intent Routes Intent routes are exit points from the AI Agent. Each route appears as a child “route node” on the canvas that you can connect to any next step (for example: a Talk to Agent block, a booking flow, or a lead capture flow). Each route includes:
   • Route Name: A clear, unique label (for example: “Talk to Sales”).
   • Description (for AI detection): A short explanation of when this route should trigger.
   • Required Variables: A list of attributes the agent must collect before triggering the route.
   Required variables map to your workspace attributes. When a route triggers, the collected values are saved back to the contact, so you can use them in later blocks.

​

How Routing Works

• The agent evaluates your routes on every user message.
• When a route triggers, the automation transitions to that route node and immediately continues to whatever block you connected after it.
• If no route triggers, the agent stays active and waits for the next user message.

​

Common Use Cases

1. AI-first support with escalation Let the agent answer from your knowledge base, then route to Talk to Agent when the user needs human help.
2. Sales qualification Route “pricing / demo / quote” intent to a route that collects lead details (name, email, budget) before starting your sales flow.
3. Routing by intent Create multiple exit routes (support, sales, order status, returns) and let the agent direct users to the right path automatically.

​

Example Use Case

• Answers product questions from a knowledge base
• Routes “Talk to Sales” requests to a sales flow
• Captures required details before routing

1. Prompt: Set a system prompt such as:
   • “You are a helpful store assistant. Answer questions using the knowledge base. If the user asks for pricing or wants to buy, route to Talk to Sales. Keep replies short and precise.”
2. Knowledge: Connect your “Products & Policies” knowledge base.
3. Routes: Add an intent route:
   • Route Name: Talk to Sales
   • Description: “Trigger when the user asks to talk to sales, asks for pricing, requests a quote, or wants to place a bulk order.”
   • Required Variables:
     • Email — “Ask for the customer’s email address.”
     • Phone — “Confirm the best phone number to reach them on.”

​

Best Practices

• Keep route names unique: Treat route names as distinct actions (for example: “Escalate to Support”, “Book a Demo”, “Talk to Sales”).
• Be explicit in route descriptions: Include common phrases users will say (“agent”, “human”, “pricing”, “quote”, “demo”, “refund”).
• Write variable prompts as natural questions: The “required variable” description should read like what the agent should ask.
• Limit routes to clear intents: Too many overlapping routes can cause ambiguous triggers.