Customizing Your AI Assistant

This comprehensive guide will help you tailor your AI assistant to best serve your specific needs, optimize its performance, and create a unique, engaging experience for your users.

Understanding Customization Options

Agent One provides multiple layers of customization:

Personality: How your assistant communicates and presents information
Knowledge: What your assistant knows and can reference
Tools: Special capabilities your assistant can use
Appearance: How your assistant is presented to users
Behavior: How your assistant handles different scenarios

Basic Customization

Agent Editor vs. Settings

In your agent dashboard, you have two main areas for configuration:

Agent Editor: This is the main workspace where you define your agent's behavior, instructions, knowledge base, and tools. Accessed via the "Agent Editor" tab in the sidebar.
Settings: This area manages your agent's identity and metadata, such as template status and category. Accessed via the "Settings" tab (gear icon) in the sidebar.

Accessing Settings

Go to your dashboard at app.agnt.one
Select the assistant you want to customize
Click "Agent Editor" to configure behavior, or "Settings" for metadata.

Creating a Template

You can share your agent setup as a template for others to use:

Navigate to the Settings tab (gear icon in the sidebar).
Locate the Template Status field and set it to "Yes".
Enter a Template Category (e.g., "Customer Support", "Education", "Productivity") to help users find your template.
Save your changes.

Your agent will now be listed in the Template Gallery for your site.

Profile Settings

In the Agent Editor, you can adjust:

Name: Choose a name that reflects your assistant's purpose or personality
Initial Message: Customize the first message users see when they start a chat
Icon: Upload a unique image to represent your assistant (recommended size: 192x192px)

SEO Description: In the "More settings" section at the bottom of the editor, you can add an SEO Description. This text is used when your agent is shared on social media or indexed by search engines.

Welcome Message Tips

Your welcome message sets the tone for the entire interaction. Consider:

Clearly stating what your assistant can help with
Suggesting starter questions users might ask
Keeping it concise but friendly

Example welcome messages:

Hi there! I'm TechBot, your technical support assistant. I can help with product troubleshooting, installation guides, and warranty information. What can I help you with today?

Welcome! I'm your travel planning assistant. I can recommend destinations, provide budget tips, or help plan itineraries. How can I assist with your travel plans?

Configuring Assistant Behavior

The "Instructions" tab is where you define how your assistant behaves.

Writing Effective Instructions

Instructions directly influence how your assistant responds to queries. They should include:

Role and Identity: Who the assistant is and what it represents
Tone and Style: How the assistant should communicate (formal, casual, technical, etc.)
Knowledge Areas: Topics the assistant should be knowledgeable about
Limitations: Topics to avoid or specific responses to certain questions
Response Format: How information should be structured

Example Instructions Template

You are [ROLE], representing [COMPANY/ORGANIZATION].

TONE:
- Maintain a [FORMAL/FRIENDLY/TECHNICAL] tone
- Use [SIMPLE/TECHNICAL] language that's easy to understand
- [INCLUDE/AVOID] humor and personality in responses

KNOWLEDGE AREAS:
- You specialize in [SPECIFIC TOPICS]
- You can help with [SPECIFIC TASKS]
- When discussing our products, emphasize [KEY FEATURES/BENEFITS]

LIMITATIONS:
- Do not provide information about [RESTRICTED TOPICS]
- If asked about pricing, [GIVE GENERAL RANGES/DIRECT TO WEBSITE/etc.]
- Never [SPECIFIC BEHAVIORS TO AVOID]

RESPONSE FORMAT:
- Keep responses [BRIEF/DETAILED] but comprehensive
- Format technical instructions as numbered steps
- Include relevant links when appropriate
- For complex topics, break information into sections with headings

COMMON SCENARIOS:
- If a user asks about [COMMON QUESTION], respond with [PREFERRED RESPONSE]
- When users need technical support, collect [SPECIFIC DETAILS] before providing solutions
- If you can't help with a query, suggest [ALTERNATIVE RESOURCES]

Best Practices for Instructions

Be specific: Vague instructions lead to inconsistent responses
Provide examples: Include sample responses for common questions
Update regularly: Refine based on actual user interactions
Test thoroughly: Try various queries to ensure instructions are effective

Knowledge Base

The knowledge base is what allows your assistant to provide accurate, specific information.

Setting Up an Effective Knowledge Base

In the "Knowledge" tab, click "Upload Files"
Select documents relevant to your assistant's purpose
Supported formats include PDF, DOCX, TXT, CSV, and more
After upload, you can organize and add context to documents

Training on Website Content

In addition to uploading files, you can train your agent by crawling a website. This is particularly useful for agents based on existing documentation or company websites.

Using the Creator Assistant:

In the Agent Editor, locate the "Agent Creator Assistant" chat (left side on desktop, or via the info button)
Type a request like: "Scrape my website https://example.com"
The assistant will open a dialog to confirm the URL
Click "Start Crawling" to process the site

The crawled content will be added to your agent's knowledge base automatically.

GitHub Repository Integration

You can import code and documentation directly from your GitHub repositories. This is perfect for creating technical assistants or coding helpers that know your specific codebase.

To add a GitHub repository:

In the Knowledge Base section, look for the repository selector (available if your GitHub account is connected).
Select one of your repositories from the dropdown list.
Click "Add GitHub Repository".
The system will ingest the repository's content and add it to your knowledge base.

Note: This feature ingests the current state of the repository. If you update the code, you may need to re-add it to refresh the knowledge.

Knowledge Base Best Practices

Chunk information logically: Break large documents into topical sections
Use descriptive filenames: Help your assistant understand what each file contains
Include comprehensive information: Cover common questions and edge cases
Keep information current: Remove outdated information and add new content as needed
Structure matters: Well-organized documents are easier for your assistant to reference

Example Knowledge Base Structure

For a product support assistant:

Product_Specifications.pdf
Installation_Guide.pdf
Troubleshooting_Common_Issues.pdf
Warranty_Information.pdf
FAQ_Compiled.pdf

For a real estate assistant:

Property_Listings_Current.pdf
Neighborhood_Guides.pdf
Mortgage_Information.pdf
Buying_Process_Overview.pdf
Rental_Policies.pdf

Email Collection

Email collection allows you to generate leads while providing valuable assistance.

Configuration Options

In the "Tools" tab, find the Email Collection section
Toggle the feature ON
Choose when to collect emails:
- Beginning: Asks for email when the conversation starts
- End: Asks for email as the conversation concludes
Customize the collection prompt to match your brand voice

Note: To view and export collected emails, see the Monitoring Your Agents guide.

Example Prompts

Beginning of conversation:

To provide you with the best assistance and follow up if needed, could you share your email address with me?

End of conversation:

I'd be happy to send you additional resources on this topic. What email address should I use?

Email Collection Tips

Explain the benefit: Tell users why sharing their email is valuable
Keep it optional: Don't force email collection if the user declines
Offer value: Provide a reason why the user would want to share their email
Respect privacy: Assure users their email will only be used for relevant communications

Custom API Functions

Custom API functions extend your assistant's capabilities by connecting to external data and services.

Setting Up API Functions

In the "Tools" tab, find the Custom API Functions section
Click "Add API Function"
Configure each API with:
- Function Name: A descriptive identifier (e.g., "check_product_inventory")
- Description: What the function does and when to use it
- Endpoint URL: The API endpoint (e.g., "https://api.yourstore.com/inventory/{product_id}")
- Parameters: Define the input parameters using JSON Schema format

Example API Configuration

For a product inventory checker:

Function Name: check_product_inventory

Description:

Checks the current inventory status of a specific product. Use when customers ask about product availability.

Endpoint URL:

https://api.yourstore.com/inventory/{product_id}

Parameters:

{
  "product_id": {
    "type": "string",
    "description": "The product's unique identifier (SKU or product code)",
    "required": true
  },
  "location_id": {
    "type": "string",
    "description": "Store location ID to check stock at a specific store",
    "required": false
  }
}

API Integration Best Practices

Public endpoints only: APIs must be publicly accessible without authentication
Simple, focused APIs: Each API should do one thing well
Thorough parameter descriptions: Help your assistant understand what each parameter means
Error handling: Ensure APIs return helpful error messages
Test comprehensively: Try different parameter combinations before deployment

Tools and Capabilities

Enable additional tools to enhance your assistant's functionality.

Code Interpreter

Ideal for assistants that need to:

Generate and explain code examples
Format data or perform calculations
Create diagrams or visualizations

To enable:

Navigate to the "Tools" tab
Toggle "Code Interpreter" to ON

File Search

Essential for knowledge base functionality:

Toggle "File Search" to ON
Ensure your knowledge base is properly configured
Your assistant will automatically search these files when answering questions

Web Search

Give your agent access to real-time information from the internet:

Toggle "Enable Web Search" to ON in the agent settings
Your agent will now be able to search for current events, facts, and live data when answering questions

Rich Media (Charts)

Visualize data directly in the chat:

Toggle "Enable Rich Media (Charts)" to ON
Ask your agent to display data (e.g., "Create a pie chart of our budget distribution")
The agent will render an interactive chart

MCP Servers

The Model Context Protocol (MCP) allows you to connect your agent to external MCP servers, enabling it to access a wide range of tools and data sources.

To add an MCP Server:

In the "Tools" tab, find the "MCP Servers" section
Click the settings icon or "Add Server" button
Enter the URL of your MCP server
Click "Save"

Your agent will now be able to discover and use tools provided by that server. This is an advanced feature for extending your agent's capabilities beyond standard APIs.

Voice Mode (Input & Output)

Agent One includes Voice Mode capabilities to make interactions more natural. Your agent can both listen to your voice input and speak its responses aloud.

For a detailed guide on using this feature, please see Using Voice Mode.

Enabling Voice Mode

To enable this feature for your agent:

Go to your agent's Settings.
Expand the More settings section at the bottom.
Toggle the Enable Voice Mode (Input & Output) checkbox.
Save your changes.

Note: This feature is available on Basic and Standard plans.

Using Voice Mode

Once enabled, you will see new controls in the chat interface:

Voice Input: Click the Microphone icon in the input bar to start speaking.
Voice Output: Click the Play button (speaker icon) next to any message to hear it read aloud.

Conversation History

Agent One automatically saves your conversations, allowing users to revisit past interactions.

Accessing Past Chats

Desktop: A sidebar on the left displays a list of your previous conversation threads. Click any thread to resume that chat.
Mobile: Tap the menu icon (hamburger menu) in the top-left corner to open the history sidebar.

Starting a New Chat

To start a fresh conversation context:

Open the conversation history sidebar.
Click the "New Chat" button (plus icon).
This creates a new thread, separate from your previous discussions.

Integration Options

You can integrate your AI assistant into your website using our embeddable chat widget:

In your agent's settings, find the "Show Embed Code" button.
Click the button to reveal the embed code.
Copy the provided JavaScript snippet.
Paste the snippet into your website's HTML, just before the closing </body> tag.

Embed Code Format

The embed code uses a simple script tag with configuration provided via data-* attributes:

<script
  src="https://app.agnt.one/widget.js"
  data-agent-id="your-agent-id"
  data-domain="https://your-site-subdomain.agnt.one"
  data-primary-color="#000000"
  data-position="bottom-right"
  data-greeting="Hi! How can I help you?"
></script>

Configuration Options

You can customize the widget by adjusting the attributes in the script tag:

data-agent-id (Required): The unique ID of your agent.
data-domain (Required): The URL where your agent is hosted (e.g., https://your-site.agnt.one).
data-primary-color: (Optional) Hex color code for the widget button (e.g., #FF5733). Defaults to black (#000000).
data-position: (Optional) Where the widget appears. Values: bottom-right (default) or bottom-left.
data-greeting: (Optional) A short welcome message that appears in a bubble above the widget button before it's opened.
data-logo: (Optional) URL to a custom logo image for the launcher button.

Widget Behavior

The chat widget has been designed to integrate seamlessly with your website:

Appearance: The widget appears as a circular launcher button.
Activation: When clicked, it expands into a chat interface.
Responsiveness: Automatically adapts to mobile and desktop views.
Context: Maintains conversation context throughout the user's session.

Full Page Integration

For a dedicated chat experience:

Your assistant is automatically available at your site URL (e.g., https://your-site.agnt.one)
You can link to this URL from your main website
Users get a full-screen chat interface optimized for longer interactions

This option is ideal for:

Dedicated support pages
Knowledge base portals
Complex customer service scenarios
Applications where the chat is the primary interface

Slack Integration

Connect your agent to a Slack workspace to allow team members to interact with it directly within Slack.

To enable Slack integration:

Open your agent's "Settings"
Expand the "More settings" section at the bottom
Look for "Slack Integration"
Click "Add to Slack"
Follow the authorization prompts to connect your workspace

Once connected, you can mention your agent in channels or DM it to get responses.

Testing and Refinement

Continuously improve your assistant through testing and observation.

User Feedback

Your agent includes a built-in feedback system to help you gauge the quality of its responses.

Thumbs Up / Down: Users can rate any response from the agent by clicking the thumbs up or thumbs down icons below the message.
Collecting Insights: This feedback is collected to help you identify which responses are helpful and which ones need improvement.

Encourage your users (or internal testers) to use these buttons so you can gather valuable data on your agent's performance.

Test Scenarios

Create a test plan covering:

Common user questions
Edge cases and unusual queries
Complex scenarios requiring multiple back-and-forth interactions
Queries that should trigger API calls
Situations where email collection should activate

Refinement Process

Review actual conversations in your dashboard
Identify patterns where your assistant could improve
Update instructions, knowledge base, or API configurations
Test changes with similar scenarios
Deploy updates when satisfied with improvements

Best Practices Summary

Clear Instructions: Be specific about how your assistant should behave
Comprehensive Knowledge: Provide well-organized, relevant information
Focused APIs: Build simple, reliable API connections
Regular Updates: Keep knowledge and capabilities current
User-Centric Design: Consider the user experience in all customizations

Troubleshooting

Assistant Not Using Knowledge Base

Verify files are in supported formats (PDF, TXT, DOCX)
Check file permissions and upload status
Ensure questions are relevant to uploaded content
Consider reformatting complex documents

API Integration Issues

Verify endpoints are publicly accessible
Check parameter configurations match expected input
Test API calls independently to ensure they work
Review error responses to pinpoint issues

Email Collection Problems

Confirm the feature is enabled
Check that the prompt is appropriate and clear
Verify the timing setting (beginning/end) matches expectations

For more detailed support, please contact our team through your dashboard or refer to the FAQs.