Skip to main content

Creating Chat Assistants

Chat assistants enable real-time, AI-powered conversations with your website visitors. This guide walks you through the process of creating, configuring, and deploying a chat assistant on your website.

Before You Begin

Before creating a chat assistant, it’s helpful to:
  • Define the primary purpose of your assistant (support, sales, information, etc.)
  • Identify common questions your assistant should answer
  • Prepare any knowledge documents your assistant might need
  • Consider the tone and personality your assistant should have

Creating a New Chat Assistant

Step 1: Initiate Creation

  1. Navigate to the Assistants section in the sidebar
  2. Click the + Create Assistant button in the top-right corner
  3. Select Chat as the assistant type
  4. Click Continue
Create Chat Assistant [Screenshot placeholder: Assistant creation screen with Chat type selected]

Step 2: Basic Configuration

In the first section of the setup wizard, configure the basic settings:
  1. Name: Give your assistant a descriptive name (e.g., “Support Bot”, “Sales Assistant”)
  2. Description: Add a brief description of the assistant’s purpose
  3. Avatar: Upload a custom avatar image or select one from the library
  4. Language: Choose the primary language for your assistant
Choose a name that reflects the assistant’s function. This helps users understand who they’re talking to.

Step 3: AI Model Configuration

Configure the AI model settings to control your assistant’s behavior:
  1. Instructions: Write detailed instructions for your assistant’s behavior, knowledge, and tone
    • Be specific about how it should respond
    • Define its role and limitations
    • Specify when it should escalate to a human
  2. Model: Select the AI model to power your assistant
    • GPT-4: Most capable model for complex conversations
    • GPT-3.5: Good balance of performance and cost
  3. Temperature: Adjust the creativity level (0.0-1.0)
    • Lower values (0.2-0.4): More consistent, conservative responses
    • Higher values (0.6-0.8): More creative, varied responses
Chat Assistant Model Settings [Screenshot placeholder: Model configuration screen with parameters]
You are a customer support representative for Acme Inc., a company that sells office supplies.

Your name is Alex, and you should be helpful, friendly, and professional.

You can help with:
- Product information and recommendations
- Order status inquiries (but you'll need an order number)
- Return policies and procedures
- General company information

If asked about order specifics without an order number, politely ask for the order number.

If users have complex issues or want to speak with a human, offer to create a support ticket or transfer them to a human agent.

Avoid:
- Making up information about products or policies
- Promising specific delivery dates
- Handling personal data like credit card information
- Discussing competitors in detail

Always maintain a positive, helpful tone, and don't use excessive technical jargon.

Step 4: Knowledge Base Connection

Connect relevant knowledge to help your assistant provide accurate information:
  1. Click the Knowledge tab in the setup wizard
  2. Select files from your Knowledge Base to associate with this assistant
  3. Adjust the knowledge retrieval settings if needed:
    • Relevance Threshold: Set how closely content must match a query (0.0-1.0)
    • Number of References: How many knowledge snippets to include per response
Knowledge Base Connection [Screenshot placeholder: Knowledge connection interface with selected files]
If you haven’t uploaded knowledge files yet, you can add them later by editing the assistant.

Step 5: Communication Settings

Configure how your assistant handles conversations:
  1. Welcome Message: Set the initial message users see when starting a conversation
  2. Fallback Response: Define what the assistant says when it can’t answer a question
  3. Conversation Timeout: Set how long until an inactive conversation closes
  4. File Sharing: Enable or disable file uploads from users
Communication Settings [Screenshot placeholder: Communication settings screen]

Step 6: Human Handoff

Configure when and how your assistant transfers conversations to human agents:
  1. Handoff Triggers: Define conditions that trigger a human handoff
    • Explicit user request
    • Multiple fallback responses
    • Specific keywords or topics
    • Sentiment detection
  2. Handoff Behavior:
    • Set notification preferences for agents
    • Configure queue behavior
    • Define handoff messages
Human Handoff Settings [Screenshot placeholder: Human handoff configuration screen]

Step 7: Advanced Settings

Configure additional options for your chat assistant:
  1. Persistent Context: How much conversation history to maintain
  2. Analytics: Enable detailed conversation analytics
  3. Message Ratings: Allow users to rate assistant responses
  4. Custom Variables: Define variables to track during conversations

Step 8: Finalize Creation

  1. Review all settings across the different tabs
  2. Click Create Assistant to save your configuration
  3. You’ll be redirected to the assistant details page

Testing Your Chat Assistant

Before deploying to your website, test your assistant:
  1. On the assistant details page, click the Test tab
  2. Start a conversation with your assistant
  3. Try different questions and scenarios to evaluate responses
  4. Check how knowledge retrieval works with various queries
  5. Test human handoff functionality if configured
Testing Chat Assistant [Screenshot placeholder: Testing interface with sample conversation]

Deploying on Your Website

To deploy your chat assistant on your website:
  1. Navigate to the Channels tab on your assistant’s page
  2. Click + Add Channel and select Website Widget
  3. Configure the widget appearance:
    • Colors and theme
    • Position on page
    • Size and behavior
  4. Configure visitor information collection if needed
  5. Click Save Configuration
  6. Copy the generated JavaScript code
  7. Paste the code into your website’s HTML, just before the closing </body> tag
Chat Widget Configuration [Screenshot placeholder: Widget configuration screen]
<!-- Flowbotic Chat Widget -->
<script>
  window.flowboticSettings = {
    widgetId: "your-widget-id",
    assistantId: "your-assistant-id",
    theme: "light",
    position: "right"
  };
  
  (function(d) {
    var s = d.createElement("script");
    s.src = "https://cdn.flowbotic.com/widget/v1/flowbotic.js";
    s.async = 1;
    d.getElementsByTagName("head")[0].appendChild(s);
  })(document);
</script>
<!-- End Flowbotic Chat Widget -->

Best Practices

Writing Effective Instructions

  • Be specific about the assistant’s role and knowledge
  • Define the tone and personality clearly
  • Include examples of good responses
  • Specify limitations and escalation criteria

Optimizing Knowledge

  • Use clear, concise documents focused on single topics
  • Break large documents into smaller, focused files
  • Include common variations of terms and questions
  • Update knowledge regularly as information changes

Testing and Refinement

  • Test with a variety of questions, including edge cases
  • Review conversation logs to identify improvement areas
  • Update instructions based on common misunderstandings
  • Add new knowledge files to address gaps

Next Steps