AI Agents

What Are AI Agents?

An AI Agent is your automated customer service representative. Unlike traditional chatbots with scripted responses, AI Agents understand natural language and can handle a wide variety of customer questions.

Every agent is powered by the ChatSpark AI Engine, our intelligent 4-step process that understands customer intent, finds the most relevant answers from your training data, applies smart reranking to select the best response, and delivers it in your brand voice. Learn more about how the AI Engine works.

Each agent is:

• Trained on your content: Knows your business, products, and policies
• Always available: Works 24/7 without breaks or downtime
• Consistently helpful: Every customer gets the same quality response
• Infinitely scalable: Handle 1 or 1,000 conversations simultaneously

Capabilities

ChatSpark AI Agents can:

• Answer questions: About your products, services, policies, and more
• Guide users: Help customers navigate your website or find information
• Capture leads: Collect contact information and qualify prospects
• Book appointments: Integrate with Calendly, Google Calendar, or Square for scheduling
• Speak multiple languages: Automatically detect and respond in 100+ languages
• Handle complex queries: Multi-part questions and follow-ups

Conversation Memory

Your AI Agent maintains context throughout a conversation session:

• Session-based memory: Remembers everything discussed in the current chat
• Contextual follow-ups: Handles "tell me more about that" naturally
• Cross-reference: Connects information from earlier in the conversation

Conversation memory persists for the duration of a chat session. If a customer returns later, it's treated as a new conversation (though you can view their full history in your dashboard).

Limitations

It's important to understand what AI Agents cannot do out of the box:

• Make decisions: Won't approve refunds or make policy exceptions
• Know everything: Limited to the training data you provide
• Handle sensitive issues: Complex complaints need human touch
• Process payments: Can't directly handle transactions

Agent Type

When creating an agent, select the type that matches your use case:

• Website: Embed a chat widget on your website (default)
• CoPilot: Create an agent optimized for internal team use via the CoPilot browser extension. Customer-facing settings like lead capture, welcome messages, and widget customization are hidden. The personality generator is tailored for internal workflows.
• Slack: Deploy as a Slack bot in your workspace
• Facebook: Connect to Facebook Messenger
• Instagram: Respond to Instagram DMs
• WhatsApp: Connect via WhatsApp Business API
• Telegram: Deploy as a Telegram bot

The agent type determines which settings and integrations are available. For customer-facing use, Website is the best starting point. For internal team productivity, choose CoPilot. See Integrations for detailed setup instructions for each channel.

Basic Settings

Configure your agent's identity and appearance:

Internal Name

A name for your reference only, displayed in your dashboard under "AI Agents." Customers never see this, so use it to organize multiple agents (e.g., "Support Bot" or "Sales Assistant - EN").

Agent Name

The public-facing name customers see in the chat widget header. Make it friendly and on-brand:

• "Alex" - Human-like and approachable
• "Support Bot" - Clear about being AI
• "Acme Assistant" - Brand-focused

Primary Color

Set the accent color for your chat widget. This affects:

• Chat bubble button
• Header background
• Send button and UI accents
• Link colors in responses

Enter any hex color code (e.g., #0083D3) to match your brand.

Agent Avatar

Upload a custom image that appears in the chat widget and next to agent responses. Recommended:

• Square image (1:1 aspect ratio)
• At least 64×64 pixels
• PNG or JPG format
• Clear on both light and dark backgrounds

Agent Personality

The Agent Personality field is where you define how your agent communicates. This is one of the most powerful customization options, shaping tone, behavior, and response style.

Write instructions as if briefing a new team member:

• Define the role: "Act as a friendly sales representative"
• Set the tone: "Be professional but approachable"
• Add boundaries: "Never discuss competitor products"
• Include behaviors: "Always ask if they need help with anything else"

Example personality instructions:

Act as a proactive sales representative for [Company Name]. Highlight product features, understand user needs to make personalized recommendations, and handle queries about pricing and promotions with clarity. Be friendly and helpful, use emojis sparingly, and always offer to connect with a human for complex questions.

Welcome Message

The welcome message is the first thing visitors see when they open the chat widget. A good welcome message:

• Greets the visitor warmly
• Explains what the agent can help with
• Invites them to ask a question

You can use rich text formatting including:

• Bold and italic text
• Bullet points and numbered lists
• Links to key pages
• Line breaks for readability

Example welcome message:

Hi there! 👋 I'm here to help you find what you need. Whether you have questions about our products, pricing, or just want to learn more, go ahead and ask!

Suggested Prompts

Suggested prompts appear as clickable buttons below the welcome message, helping visitors start conversations without typing. They're conversation starters that reduce friction.

• Display up to 5 suggested prompts
• Guide users toward common topics you want to address
• Reduce friction for customers who don't know what to ask
• Improve engagement with one-click interactions

Example prompts:

• "What are your pricing plans?"
• "How do I get started?"
• "What integrations do you support?"
• "Can I schedule a demo?"

Custom Unknown Response

When your agent encounters a question outside its training data, it needs a graceful fallback. The Custom Unknown Response lets you control exactly what your agent says in these situations.

Default behavior:

"I'm sorry, I don't have that information. How can I assist you further?"

You can customize this to match your brand voice and guide customers toward helpful next steps:

• Offer to connect them with your team
• Suggest related topics you can help with
• Provide contact information for complex inquiries
• Trigger lead capture to follow up later

Configure in your agent settings under Settings → Agent Unknown Response.

Lead Capture

ChatSpark's built-in Lead Capture collects customer contact information through natural conversation, either automatically when the chat opens or triggered by specific keywords. Captured leads can be exported as CSV or pushed anywhere via Zapier.

• Configurable fields: Name, email, phone, company, and message
• Automatic or trigger-based: Capture on load or when keywords are detected
• Custom messaging: Personalize request, completion, and follow-up messages
• Zapier integration: Push to CRMs, email lists, or create support tickets, and more
• CSV export: Download leads directly from Chat History

Configure Lead Capture in your agent's Settings tab under Lead Capture.

Canned Responses

Canned Responses let you define specific messages that trigger when customers use certain words or phrases. This is useful for routing conversations or providing instant answers to common requests.

Common use cases:

• Live agent routing: When someone says "speak to a person" or "live agent," display contact options or initiate handoff
• Appointment booking: Trigger your booking flow when someone mentions "schedule" or "book appointment"
• Special promotions: Share current offers when keywords like "discount" or "promo code" are detected
• Emergency contacts: Immediately provide urgent contact info for time-sensitive situations

Configure Canned Responses in your agent settings under Settings → Canned Responses. Add trigger words and the message you want displayed.

Content Carousel

For e-commerce and product-focused businesses, Content Carousel displays rich, visual cards within chat responses. Instead of plain text, your agent can showcase products or services with images and descriptions.

• Visual product cards: Display images, titles, and descriptions
• Service showcases: Present offerings in an engaging format
• Better engagement: Visual content drives higher interaction
• Mobile-optimized: Swipeable cards work great on all devices

Enable Content Carousel in your agent settings under Settings → Content Carousel.

Handoff to Humans

ChatSpark AI Agents handle the vast majority of conversations autonomously. For businesses that need human backup for specific situations, there are several escalation options:

• Live agent handoff via AI Actions: Transfer conversations in real-time to Freshchat, Intercom, or HappyFox Chat with full conversation context
• Support ticket creation via AI Actions: Automatically open tickets in Zendesk, Freshdesk,HappyFox, or Intercom with conversation context attached
• Lead capture: Collect contact info so your team can follow up asynchronously
• Email transcripts: Send full conversation history to your team for review
• Zapier: Trigger any workflow when handoff conditions are met

Google Analytics Tracking

Connect your AI agent to Google Analytics 4 (GA4) to track AI agent interactions alongside your other website metrics. This is essential for KPI tracking, measuring ROI, and understanding customer behavior, especially for enterprise teams reporting on support automation metrics.

Events tracked include:

• Chat session starts: When a visitor opens the chat widget
• Messages sent by users: Every customer message
• Bot responses: Every agent reply
• Link clicks from bot responses: Clicks on URLs in agent messages
• Lead capture events: When contact information is collected
• Suggested prompt clicks: Which prompts visitors click
• Chat session ends: When conversations close

To enable:

1. Get your GA4 Measurement ID from Admin → Data Streams in Google Analytics
2. Enter the ID in your agent settings under Settings → Google Analytics Integration
3. Events will begin appearing in your GA4 reports