Out-of-the-box in: UI Kit Builder ✅ | Widget Builder ✅ | UI Kits ✅ | SDK ✅
Prerequisites
Open your CometChat app
Login to your CometChat dashboard and choose your app.
Configure AI
Set the GPT model
Enter the OpenAI ChatGPT model name you intend to use (e.g.,
gpt-4, gpt-3.5-turbo).Save the OpenAI key
Generate an API key from your OpenAI account and save it here.
Set a custom instruction
Add app-level context to guide how CometChat AI responds. Example: “You are a helpful customer support assistant for an e-commerce platform.”
Temperature Guidelines
| Temperature | Behavior | Best for |
|---|---|---|
| 0.0 - 0.3 | Highly consistent, deterministic | Customer support, factual responses |
| 0.4 - 0.6 | Balanced | General conversation, most use cases |
| 0.7 - 1.0 | Creative, varied | Creative writing, brainstorming |
Start with 0.3-0.5 for most business applications. Increase if responses feel too repetitive.
Conversation Starter
Conversation Starter enables you to retrieve an initial message in a new conversation, often used to set the context for the conversation that is about to begin. When to use: Empty chat screens, onboarding new users, or helping users break the ice in new conversations.Before you begin
Enable Conversation Starter
Go to Chat → Features, under AI User Copilot, enable Conversation Starter.
Use v4 Chat SDKs
Implement chat using CometChat v4 Chat SDKs.
How does it work?
CometChat AI analyzes the user’s tone and writing style by reviewing recent messages sent by that user within the application. The SDK includes a method for retrieving conversation starters in a chat. This method returns an array containing three potential starters for the conversation. Configuration options:- lastNMessages: Fetch a specific number of recent messages.
- fromTimestamp: Fetch messages from a particular timestamp.
- toTimestamp: Fetch messages up to a particular timestamp.
A maximum of only
1000 messages will be fetched regardless of configuration.UI Kits
Conversation Starter functions seamlessly starting from v4 of the Chat UI Kits. It is triggered automatically when there are no messages in a conversation.Smart Replies
Smart Replies enable the retrieval of an AI-generated response message within a conversation. Helps users respond quickly with contextually appropriate messages. When to use: Busy support agents, users who want quick responses, or reducing typing effort on mobile.Before you begin
Use v4 Chat SDKs
Implement chat using CometChat v4 Chat SDKs.
How does it work?
CometChat AI goes through the messages of a conversation to understand the context & provide relevant replies. It returns three replies:positive, negative & neutral.
Configuration options:
- lastNMessages: Fetch a specific number of recent messages.
- fromTimestamp: Fetch messages from a particular timestamp.
- toTimestamp: Fetch messages up to a particular timestamp.
- unreadOnly: Fetch only unread messages.
A maximum of only
1000 messages will be fetched regardless of configuration.UI Kits
Smart Replies function seamlessly starting from v4 of the Chat UI Kits. In v4, Smart Replies are accessible manually, whereas in v5, they work automatically.Conversation Summary
Conversation Summary enables the summarization of conversations using AI. Get quick overviews of long conversations without reading every message. When to use: Catching up on missed conversations, handoffs between support agents, or reviewing discussion outcomes.Before you begin
Enable Conversation Summary
Go to Chat → Features, under AI User Copilot, enable Conversation Summary.
Use v4 Chat SDKs
Implement chat using CometChat v4 Chat SDKs.
How does it work?
CometChat AI goes through the messages of a conversation to understand the context & provide a short summary of the conversation. The SDK method returns the conversation summary as a string. Configuration options:- lastNMessages: Fetch a specific number of recent messages.
- fromTimestamp: Fetch messages from a particular timestamp.
- toTimestamp: Fetch messages up to a particular timestamp.
- unreadOnly: Fetch only unread messages.
A maximum of only
1000 messages will be fetched regardless of configuration.