Skip to main content

Create a Chat Widget

Chat Widgets are embeddable AI assistants that integrate directly into your website or application. They offer fast deployment, domain-level access controls, and lightweight customization for a seamless user experience.

Use Chat Widgets for documentation sites, in-app help, marketing pages, or anywhere users need instant answers from your data.

Prerequisites​

Before creating a Chat Widget, ensure you have:

  • An active Vectorize account with a paid plan (Starter or higher)
  • At least one pipeline created with indexed documents
  • A website or application where you'll embed the widget

What You'll Build​

By the end of this guide, you'll have:

  • An embeddable widget powered by your RAG pipeline
  • Custom authentication and domain restrictions
  • A configured AI model with defined behavior
  • Integration code for use on your site

1. Create a New Agent​

Step 1: Navigate to Agents​

From the Vectorize platform, click Agents in the sidebar.

Step 2: Click “New Agent”​

Use the button in the top-right corner.

2. Configure Your Widget​

Agent Name​

Enter a name (3–50 characters) for internal reference.

Description (Optional)​

Add a short note for your team—this is not visible to end users.

Application Type​

Select Widget.

Active Status​

Choose whether to activate the widget immediately. You can enable or disable it later.

Agent Basic Info

3. Select a Model​

Widgets use a single AI model for consistency and simplicity.

Choose a Model​

  1. Select your platform:
    • Built-in (no key required)
    • Connected platforms (OpenAI, Anthropic, Groq)
  2. Choose a model from the list
  3. If needed, click Add LLM Provider to add new integrations

The selected model will be used for all widget interactions.

4. Set the System Prompt​

The system prompt defines your assistant’s behavior and tone.

Prompt Tips​

  • Be concise—space is limited
  • Reference your website context
  • Provide a fallback (“check our docs”)

Example:

You are a helpful assistant embedded on [website name].
Answer questions about our product and documentation.
Keep responses short. If unsure, suggest contacting support at [URL].

Model Parameters​

  • Temperature (0–2):
    • 0.3–0.5 for accuracy
    • 0.7 default
  • Max Tokens:
    • 500–1000 recommended
    • 2048 max

5. Set Authentication​

Widgets require admin credentials for secure management.

Admin Credentials​

  • Username: Unique identifier (e.g. widget-support)
  • Password:
    • At least 8 characters
    • Must include letters, numbers, and symbols
  • Confirm Password: To prevent typos

These credentials are hashed and used for admin access and secure API calls.

6. Configure CORS Settings​

CORS (Cross-Origin Resource Sharing) ensures only approved domains can embed the widget.

Allowed Origins​

List one domain per line. Examples:

https://example.com
https://www.example.com
http://localhost:3000

Avoid wildcards (*) in production. List exact domains to ensure security.

CORS Configuration

7. Create the Widget​

Review your configuration and click Create Agent.

8. Connect to Your Pipeline​

After creation, connect your widget to data via Agent Tools.

Option A: Create a New Tool​

  1. Click Create New Tool
  2. Set:
    • Tool name
    • Description
    • Pipeline
    • Optional filters

Option B: Use Existing Tool​

  1. Click Add Existing Tools
  2. Select from tools in your workspace

Your widget goes live once tools are connected.

Agent Deployed

9. Embed the Widget​

Get Your Embed Code​

  1. Go to Launch Dashboard, then to the agent's Integration tab
  2. Copy the provided code

Integration Tab

Integration Options​

  • JavaScript Snippet: Paste into your HTML (</body>)
  • NPM Package: For React, Vue, Angular apps
  • Direct API: For fully custom implementations

10. Customize the Appearance​

Built-in Options​

  • Color scheme
  • Position (bottom-left / bottom-right)
  • Size and spacing
  • Button text

Custom CSS​

Override styles using your own CSS. Target widget classes as needed.

11. Test the Widget​

Use the Admin Panel to validate setup.

Preview Mode​

  • Go to Setup Widget
  • Click Test Widget
  • Confirm model responses, tool execution, and styling

Debug Mode​

Enable debug logging in the browser console for troubleshooting.

12. Monitor & Manage​

Use the Admin Panel to configure:

  • System prompt and model behavior
  • CORS and authentication settings
  • Appearance and language
  • Webhooks and error logging

Advanced Options​

Programmatic Control​

Widgets support:

  • Opening/closing via code
  • Sending messages
  • Listening to events
  • Updating configuration dynamically

Contextual Activation​

Show widgets based on:

  • Page URL
  • Scroll position
  • Time delay
  • Custom conditions

Multi-language Support​

Supports 11+ languages. Set during initialization or based on user preference.

Next Steps​

Was this page helpful?