Tool Chaining

Hunch supports intelligent multi-tool chaining, allowing the AI to automatically execute multiple tools in sequence based on user intent, with results from one tool flowing into the next.

How Tool Chaining Works

When a user request requires multiple actions, the AI can:

1. Detect chaining intent - Recognize when multiple actions are needed
2. Execute sequentially - Run tools in the correct order
3. Pass results - Use output from one tool as input to the next
4. Synthesize response - Provide a unified response with all results

Examples

Shopping Flow

User: "Find a red sweater and add it to my cart"
↓
AI: Executes search_products → add_to_cart
↓
Result: "Found the red sweater! I've added it to your cart."

Booking Flow

User: "Check if you have the blue shirt in size M, and if so, book an appointment"
↓
AI: Executes check_inventory → book_calendly_appointment
↓
Result: "Yes, the blue shirt is available! I've booked your appointment."

Newsletter + Lead

User: "Subscribe me to the newsletter and notify your team"
↓
AI: Executes subscribe_newsletter → send_lead_notification
↓
Result: "You're subscribed! Our team has been notified."

Context Sharing

Tools automatically share context through the Tool State Manager:

Cart State

• Products added to cart persist across tools
• Cart can be checked at any point
• Checkout tool can access full cart contents

Previous Results

• Each tool's result is stored
• Subsequent tools can reference prior results
• Context is preserved for the session

User Context

• Email, name, and ID available to all tools
• Tools can personalize based on user data
• Context updates as user provides information

Supported Chains

E-commerce

Chain	Tools	Description
Search → Add to Cart	search_products, add_to_cart	Find and purchase
Add → Checkout	add_to_cart, get_cart, checkout	Complete purchase
Inventory → Book	check_shopify_inventory, book_calendly_appointment	Check availability and book

Scheduling

Chain	Tools	Description
Check → Book	check_calendly_availability, book_calendly_appointment	Find and book
Search → Schedule	search_products, book_calendly_appointment	Product inquiry to booking

Lead Capture

Chain	Tools	Description
Subscribe → Notify	subscribe_newsletter, send_lead_notification	Newsletter with alert
Contact → CRM	submit_contact_form, create_crm_lead	Form submission pipeline

Implementation

Frontend Integration

The embed widget automatically handles tool chaining - no additional configuration needed.

API Integration

For custom implementations, use the /llm/tool-chain endpoint:

POST /api/llm/tool-chain
{
  "website_id": "abc123",
  "session_id": "session_xyz",
  "messages": [
    { "role": "user", "content": "Find red sweater and add to cart" }
  ],
  "tools": [...],
  "tool_results": [
    { "toolName": "search_products", "result": { "products": [...] } }
  ],
  "continue_chain": true
}

Configuration

Chain Length Limits

• Default: 5 tools per chain
• Configurable per plan
• Prevents infinite loops

Context Injection

Results are automatically injected into subsequent tools.

Critical Tool Handling

If a critical tool (checkout, payment, booking) fails, the chain stops.

Best Practices

1. Enable relevant tools - More tools = better chaining
2. Test flows - Verify chains work for common user journeys
3. Monitor usage - Track which chains are most popular
4. Handle failures - Provide fallback options when chains fail