Creating AI Tools

Give your AI assistant superpowers with custom tools

AI SmartTalk's SmartFlow visual node builder allows you to create custom tools for your AI assistant without writing code. These tools extend your AI's capabilities beyond conversation into real actions.

---

What Are AI Tools?

Tools (also called "function calling") enable your AI to perform specific tasks during conversations. Instead of just answering questions, your AI can:

• Search your product catalog
• Book appointments
• Create support tickets
• Look up order status
• And much more

Without vs With Tools

Scenario	Without Tools	With Tools
Order Status	"Please visit our website to check your order"	AI fetches tracking info and displays it instantly
Appointment Booking	"Call us at 555-1234 to schedule"	AI checks availability and books the slot
Product Search	"Browse our catalog at..."	AI searches and shows matching products

---

Creating Tools with SmartFlow

Use the Conversation Tool trigger to create tools your AI can use during chats.

Step 1: Create a New SmartFlow

1. Go to Automation → SmartFlows
2. Click + New SmartFlow
3. Give it a descriptive name (e.g., "Product Search Tool")

Step 2: Add the Conversation Tool Trigger

Drag the Conversation Tool trigger onto the canvas:

Configure the trigger:

Field	Description
Name	Tool name the AI will recognize
Description	Helps AI understand when to use this tool
User Targeting	Who can use this tool (all users, connected only, etc.)
Tool Variables	Parameters the AI should collect from the user

Step 3: Define Variables

Add variables that your AI should collect conversationally:

Variable	Type	Description
productName	String	What product to search for
maxPrice	Number	Maximum price filter
category	String	Product category

The AI will naturally ask users for these values during conversation.

Step 4: Add Actions

Connect actions to perform the actual work:

Common action combinations:

Use Case	Actions
Product Search	Search Knowledge → Set Observation
Ticket Creation	JIRA Create Issue → Set Observation
Appointment	API Call (calendar) → Send Message
Order Lookup	API Call (e-commerce) → Set Observation

Step 5: Return Results with Set Observation

Use Set Observation to send results back to the AI:

The AI receives the observation and uses it to respond to the user.

Step 6: Test Your Tool

Click ▶ Testing to try your tool:

1. Create a test conversation
2. Type a message that should trigger your tool
3. Watch the flow execute
4. Verify the AI responds correctly

---

Example: Product Search Tool

Goal: Let users search products by asking naturally

Trigger Configuration:

• Name: searchProducts
• Description: "Search the product catalog by name, category, or price"
• Variables: query (String), maxPrice (Number, optional)

Flow:

Conversation Tool Trigger
    ↓
Search Knowledge Base
  → Query: {{trigger.query}}
  → Limit: 5 results
    ↓
Set Observation
  → Results: {{searchResults}}

User Experience:

User: Do you have any blue widgets under €50?

AI: Let me search for that...

    I found 3 blue widgets under €50:
    1. Blue Widget Basic - €29.99
    2. Blue Widget Pro - €45.00
    3. Blue Widget Mini - €19.99
    
    Would you like more details on any of these?

---

Example: JIRA Ticket Creation

Goal: Let users create support tickets from chat

Trigger Configuration:

• Name: createTicket
• Description: "Create a support ticket for technical issues"
• Variables: title (String), description (String), priority (String)

Flow:

Conversation Tool Trigger
    ↓
JIRA Create Issue
  → Project: SUPPORT
  → Type: Bug
  → Summary: {{trigger.title}}
  → Description: {{trigger.description}}
    ↓
Set Observation
  → "Ticket {{jiraIssue.key}} created successfully"

---

Best Practices

Clear Tool Descriptions

Write descriptions that help the AI understand when to use each tool:

// Good
"Search the product catalog by name, category, or price range"

// Too vague
"Search products"

Meaningful Variable Names

Use descriptive variable names and descriptions:

Variable	Description
customerEmail	Email address to look up the order
appointmentDate	Preferred date for the appointment

Handle Errors Gracefully

Include error handling in your observations:

// Success
{"found": true, "results": [...]}

// No results
{"found": false, "message": "No products match your search"}

Test Conversationally

Test your tools by chatting naturally, not just clicking buttons. Make sure the AI:

• Asks for required variables naturally
• Uses the tool at the right moment
• Presents results clearly

---

Related Documentation

• SmartFlow Introduction — Learn the node builder
• Conversation Tool Trigger — Full trigger reference
• Observations — Returning data to AI
• Actions Overview — All available actions