Skip to content
Pluggi Docs

Building an LLM-Powered Bot

What is an LLM Bot?

Section titled “What is an LLM Bot?”

An LLM (Large Language Model) Bot is an AI-powered conversational agent that uses artificial intelligence to understand customer questions and provide natural, context-aware responses. Unlike traditional rule-based bots that follow predetermined paths, LLM bots can handle complex, open-ended conversations and adapt their responses based on customer interactions.

Benefits of LLM Bots:

  • Handle complex customer questions without predefined responses
  • Provide more natural, human-like conversations
  • Reduce need for extensive flow configuration
  • Continuously improve through interactions
  • Personalize responses based on customer data

Prerequisites

Section titled “Prerequisites”

Before building an LLM-powered bot, ensure you have:

  1. Account Access - An active admin portal account (contact pluggi@support.com to register)
  2. Bot Access - Support team assigned you a bot to work with in the dashboard
  3. LLM Permissions - Request advanced LLM capabilities in the assigned bot from support
  4. Basic Knowledge - Understanding of Flow Builder and Blocks

Getting Started

Section titled “Getting Started”

Step 1: Account Access & Permissions

Section titled “Step 1: Account Access & Permissions”
  1. Register for the admin portal by contacting Pluggi support at pluggi@support.com
  2. Log in and select your assigned bot from the dashboard to begin exploration
  3. Ensure you have requested LLM permissions from support to enable advanced sales assistant capabilities

Building Your LLM Bot: Three Phases

Section titled “Building Your LLM Bot: Three Phases”

Follow these three phases to build a fully functional LLM-powered bot for your business.

Phase 1: API Inventory Integration

Section titled “Phase 1: API Inventory Integration”

Connect your inventory system so the bot can provide accurate product information.

Steps:

  1. Navigate to Settings → Integrations → API Integrations
  2. Select your POS partner from the available list:
    • Dutchie
    • Jane
    • Buddy
  3. Click Integrate and enter your API credentials provided by your POS partner
  4. Click Test Connection to verify that Pluggi can communicate with your shop
  5. Upon success, save the settings to trigger the initial data synchronization

[Figure 1: API Integration Dashboard and Sync Status]

What happens next: Your inventory data is now synchronized with Pluggi. The system will continue to update product information based on your POS system.

Phase 2: AI Assistant Configuration

Section titled “Phase 2: AI Assistant Configuration”

Set up your bot’s personality, tone, and knowledge base.

Steps:

  1. Go to Settings → Integrations → AI Assistant and enable the LLM Switch
  2. Define Behavior Instructions to set the agent’s tone and persona
    • Include details about your retail vision and core brand values
    • Specify focused services and unique selling points
    • Ensure the AI represents your brand voice accurately
  3. Upload a Knowledge Base (.txt or .docx file) containing:
    • Comprehensive business details
    • Extensive product catalogs and descriptions
    • Brand background and history
    • Delivery protocols and operating hours
    • Any other relevant business information
  4. Configure the FAQ Section with specific question-and-answer pairs
    • Use this for critical inquiries that require exact responses
    • Ensure strict adherence to specific facts (prices, policies, etc.)

[Figure 2: AI Assistant Training and Knowledge Upload Interface]

Tips for best results:

  • Provide detailed, comprehensive knowledge base materials
  • Include real examples of how the bot should respond
  • Update FAQs regularly with new customer inquiries
  • Test responses frequently and refine instructions as needed

Phase 3: Conversational Flow Design

Section titled “Phase 3: Conversational Flow Design”

Create the conversation flow that integrates the LLM block.

Steps:

  1. Open the Flow Menu and locate the default Greetings template
  2. Drag the LLM Block onto the canvas
  3. Connect the LLM Block to the Greetings Message Block
  4. Configure the LLM Block with:
    • Instructions: Specific prompts for how the Agent should respond
    • Fallback Message: A response to use if the LLM cannot answer
  5. Click Test Agent to instantly verify the conversational logic

Enhance your flow:

[Figure 3: Flow Builder Canvas showing LLM Block connection]

Best Practices for LLM Bots

Section titled “Best Practices for LLM Bots”

Training & Knowledge Base

Section titled “Training & Knowledge Base”
  • Comprehensive: Include all product details, policies, and procedures
  • Current: Keep information updated as your business changes
  • Contextual: Add examples of how the bot should handle different scenarios
  • Accurate: Verify all information before uploading

Behavior Instructions

Section titled “Behavior Instructions”
  • Clear Tone: Define exactly how you want the bot to sound
  • Brand Voice: Ensure consistency with your brand personality
  • Scope: Clearly define what the bot can and cannot help with
  • Limitations: Be transparent about bot capabilities to set customer expectations

Testing & Optimization

Section titled “Testing & Optimization”
  • Test Thoroughly: Try various customer questions before launching
  • Monitor Performance: Review actual conversations to improve responses
  • Update FAQs: Add new Q&A pairs based on real customer inquiries
  • Gather Feedback: Use customer interactions to refine the knowledge base

Customer Experience

Section titled “Customer Experience”
  • Keep it Simple: Avoid overwhelming customers with too many options
  • Be Helpful: Provide clear answers or escalation paths
  • Stay Professional: Maintain appropriate tone in all interactions
  • Handle Errors Gracefully: Plan for when the bot doesn’t understand

Testing Your Bot

Section titled “Testing Your Bot”

Before deploying your LLM bot, thoroughly test all conversation paths:

  1. Ask Complex Questions - Test with open-ended questions that require reasoning
  2. Test Edge Cases - Try unusual questions or requests
  3. Verify Product Information - Confirm that product recommendations are accurate
  4. Check Fallbacks - Ensure fallback messages trigger appropriately
  5. Test Flow Branches - Verify that all conditional paths work correctly
  6. Review Responses - Check that responses align with your brand voice

Common Challenges & Solutions

Section titled “Common Challenges & Solutions”

Challenge: Bot gives generic or unhelpful responses

Section titled “Challenge: Bot gives generic or unhelpful responses”

Solution: Enhance your knowledge base with more detailed information and specific examples. Update behavior instructions with more context about your business.

Challenge: Bot recommends out-of-stock products

Section titled “Challenge: Bot recommends out-of-stock products”

Solution: Ensure your API integration is properly configured and inventory is syncing correctly. Check that your POS system credentials are valid.

Challenge: Bot doesn’t understand customer intent

Section titled “Challenge: Bot doesn’t understand customer intent”

Solution: Add more FAQ examples covering similar questions. Update behavior instructions to be more specific about expected customer inquiries.

Challenge: Responses don’t match brand voice

Section titled “Challenge: Responses don’t match brand voice”

Solution: Review and update behavior instructions with more specific tone and personality guidelines. Test different instruction versions to find the best fit.

Next Steps

Section titled “Next Steps”

Your LLM bot is now ready! Here’s what to do next:

  1. Monitor Analytics - Check Analytics to see how customers interact with your bot
  2. Review Conversations - Use Messages to read actual customer conversations
  3. Gather Feedback - Collect customer feedback to improve responses
  4. Refine Knowledge - Update your knowledge base based on real interactions
  5. Expand Capabilities - Add more templates and flows to handle additional customer needs
  6. Publish & Deploy - When ready, publish your bot and embed it on your website

Getting Help

Section titled “Getting Help”

For technical support or questions about your LLM bot:

Section titled “Related Resources”