How to Configure Smart Agent in a Voice Flow

Learn how to add and configure a Smart Agent node within your voice flow to enable AI-driven caller interactions.

---

Overview

The Smart Agent node is the AI brain of your voice flow. It listens to what the caller says, understands their intent using a knowledge base, and generates an intelligent spoken response — all in real time.

A typical voice flow with Smart Agent follows this pattern:

Answer Call
  └── CONVERT: Text To Speak (Welcome prompt)
        └── LISTEN: Collect Speech Response (Capture caller input)
              └── NLP: Smart Agent (Understand and respond)
                    ├── CONDITION (Route based on response)
                    └── WRONG INPUT (Handle unrecognised input)

---

Step 1: Add the Smart Agent Node to the Canvas

1. Open your voice flow in the Flow Playground
2. From the left sidebar, click NLP to expand the node options
3. Drag the Smart Agent sub-node onto the canvas
4. Connect it to the preceding LISTEN node and the following CONDITION or CONVERT node
5. Click the settings icon on the Smart Agent node to open the configuration panel

The configuration panel has two tabs — General Settings and Advanced Settings.

note

The Smart Agent node will show a WRONG INPUT output path by default. Always connect this to a fallback node such as a CONVERT: Text To Speak that re-prompts the caller.

---

Step 2: Configure General Settings

General Settings define how the Smart Agent processes caller input and generates responses.

AI Provider Settings

Field	Description	Required
Provider	Select the AI service provider powering the Smart Agent	Yes
Temperature	Controls response creativity. Lower values (e.g. 0.05) give more precise, consistent answers. Higher values allow more varied responses.	No
Output Token	Set the maximum length of the AI-generated response	No

Knowledge Base Settings

Field	Description	Required
Knowledge Base	Select the knowledge base the Smart Agent will query for answers	Yes
Store KB Response In	Define the variable where the knowledge base response is stored	Yes

Toggles

Toggle	Description
Streaming	Enables real-time streaming of the AI response as it is generated. When enabled, the response streams directly to the caller — no TTS node is needed as the next step. When disabled, you must connect a CONVERT: Text To Speak node after the Smart Agent to convert the response into spoken audio.
Guard	Activates a safety layer to filter inappropriate or out-of-scope responses

Prompt and Context

Field	Description	Required
System Prompt	Define the AI's behaviour, role, and boundaries for the conversation	Yes
Context	Provide additional background information the agent should be aware of	No
User Message	Define the message or input passed to the AI from the caller's captured response	No
Store Agent Response In	Define the variable where the full AI response is stored for use later in the flow	Yes

Additional Settings

Field	Description
Skip Tag Parsing	Toggle to skip parsing of response tags in the AI output
Trigger Keyword	Define keywords that activate or trigger the Smart Agent node
Skip TTS Tag	Select tags to exclude from Text To Speak conversion in the agent's response
Event Name	Name of the event to fire upon Smart Agent completion
Event Timeout	Time in seconds before the event times out
Event Time	Scheduled time associated with the event

tip

Write a clear and specific System Prompt to define the agent's role. For example: "You are a helpful insurance assistant. Answer only questions related to claims and policies. Keep responses concise and suitable for voice."

---

Step 3: Configure Advanced Settings

Advanced Settings connect the Smart Agent to external tools and services via MCP, and configure link generation for dynamic responses.

MCP Integration

Field	Description	Required
Enable	Toggle to activate the MCP integration for this Smart Agent node	No
MCP Server	Select the MCP server the Smart Agent will connect to for tool execution	Yes (if enabled)
MCP Tools	Select the specific tools available to the Smart Agent from the MCP server	Yes (if enabled)

Click FETCH after selecting an MCP Server to load the available tools from that server.

Link Generator

Field	Description	Required
Enable Link Generator	Toggle to allow the Smart Agent to generate dynamic links in responses	No
Base Path	Define the base URL path used when generating links	Yes (if enabled)
Short Code	Define the short code appended to the base path for generated links	Yes (if enabled)

note

MCP integration allows the Smart Agent to call external tools and APIs in real time during the conversation — such as looking up a policy, retrieving claim status, or fetching account details dynamically.

---

Step 4: Connect and Save

Once both General and Advanced Settings are configured:

1. Close the configuration panel
2. Ensure the Smart Agent node is connected to the correct next nodes on the canvas:
   • Success path → CONDITION or CONVERT node to continue the flow
   • WRONG INPUT path → CONVERT: Text To Speak to re-prompt the caller
3. Click SAVE in the top bar to save your changes
4. Run a TEST CALL to validate the Smart Agent's responses before going live

---

Best Practices

• Keep the System Prompt focused and role-specific — avoid broad or open-ended instructions
• Set a low Temperature (0.05–0.2) for voice flows to ensure consistent, accurate responses
• Always handle the WRONG INPUT path to avoid dead ends in the call flow
• Use Store Agent Response In to pass the AI response to downstream nodes such as CONDITION or TRANSFER
• Test the Smart Agent with a variety of caller inputs including edge cases and unclear phrases
• Use Guard to prevent the agent from responding to out-of-scope or inappropriate requests

---

Related Articles

• Flow Overview
• How to Create a Voice Flow
• Managing Channels
• Understanding Smart Agents