AI Chat

The TM AI Chat feature provides an interactive conversational AI interface that allows users to engage with AI agents for various tasks. It supports role-based access control, contextual conversations, and optional integration with Vector Database for knowledge-enhanced responses.

Overview

TM AI Chat enables users to:

Interact with configured AI agents through a conversational interface
Maintain chat history and context across sessions
Add document data as context for more informed responses
Leverage Vector Database knowledge when available
Control access to data by configuring up-front the TM AI Chat Agent

Accessing AI Chat

Navigate to the AI Chat interface through:

Go to the AI Workspace
Click on Go to AI Chat under the AI Chat section
Or directly access via the page route: /app/tm-ai-chat

Creating a TM AI Chat Agent

AI Chat Agents define the behavior, context, and constraints for conversational AI interactions.

Agent Configuration Fields

Field	Description
Name	Unique identifier for the chat agent (e.g., "Customer Support Assistant", "HR Helper").
AI Model	The LLM model that powers this chat agent. Only models with type "LLM" are compatible.
Application	The application scope this agent belongs to. Required field that cannot be changed after creation.
Active	Toggle to enable/disable the agent. Inactive agents won't appear in the chat interface.

Chat Configuration Fields

Field	Description
Instructions	System prompt that defines the agent's personality, behavior, and capabilities. This is the primary way to control how the agent responds.
Context	Multi-select field to attach reusable context sets from TM AI Context. These provide additional background information to the agent.
Allowed Roles	Specify which user roles can access this chat agent. If empty, all Desk users can use it.
Max Chat Length	Limits the number of user messages per chat session (default: 100). When reached, users must start a new conversation.

Search Knowledge Fields (if Vector DB enabled)

Field	Description
Use Vector DB	Enable to allow the agent to search and retrieve information from the Vector Database.
Search Scope	Optional link to a VectorDB Search Scope that defines search parameters and filters. If not specified, searches the entire knowledge base.

Using the Chat Interface

Starting a Conversation

Select an Agent: Use the autocomplete field or browse available agents
Send Messages: Type your message and press Enter or click Send
View Responses: AI responses stream in real-time

Chat Features

Adding Document Context

During a conversation, you can add document data as context:

Click the Add Context button (+ icon)
Select the DocType you want to reference
Choose specific documents to include
The selected data becomes part of the conversation context

Chat History

Recent conversations appear in the sidebar (last 5 chats)
Click on any previous chat to continue the conversation
Completed chats are read-only but can be viewed
Access full history through the "See all history" link

Session Management

Each chat maintains its own context and message history
Chats have three states:
- Active: Ongoing conversation
- Completed: Reached max length or manually ended
- Archived: Old conversations for reference
Use the New Chat button to start fresh with any agent

Real-time Streaming

Responses stream character by character for immediate feedback
Loading indicators show when the AI is processing

TM AI Chat History

The system automatically maintains conversation history for each user.

History Fields

Field	Description
Chat Agent	The agent used for this conversation
User	The user who initiated the chat
Status	Current state (Active/Completed/Archived)
Messages	JSON array containing the full conversation
Last Message Time	Timestamp of the most recent interaction

Managing History

History records are created automatically when starting new chats
Users can only view their own chat history
Administrators with TM AI Chat Admin role can view all histories
Old conversations can be archived for long-term storage

TM AI Chat Settings

Global settings control the availability of the chat feature.

Field	Description
Enabled	Master toggle to enable/disable the entire chat feature system-wide

Permissions and Roles

The chat system uses two primary roles:

TM AI Chat Admin

Full access to create, edit, and delete chat agents
Can view all chat histories across users
Manages global chat settings
Can delete chat history records

TM AI Chat User

Can view available chat agents (read-only)
Access their own chat history
Cannot modify agent configurations
Cannot access other users' conversations

Best Practices

Agent Design

Clear Instructions: Write specific, detailed instructions that clearly define the agent's role and boundaries
Contextual Knowledge: Use TM AI Context to provide domain-specific information without repeating it in instructions
Role Specificity: Name agents by their function (e.g., "Legal Advisor", "Code Reviewer") for clarity
Access Control: Use role restrictions for sensitive or specialized agents

Conversation Management

Session Length: Set appropriate max chat lengths based on use case (shorter for Q&A, longer for analysis)
Context Usage: Add document context only when relevant to avoid information overload
History Review: Periodically review chat histories to understand usage patterns and improve agents

Performance Optimization

Vector DB Integration: Enable for agents that need to search large knowledge bases
Search Scopes: Use specific search scopes to improve relevance and control data access
Context Limits: Balance between comprehensive context and response time

Integration with Vector DB

When Vector DB is enabled:

Chat agents can search the knowledge base for relevant information
Search results are automatically included as context
Multi-turn conversations can use query rewriting for better results
Search scopes allow fine-tuned control over what information is accessible

Troubleshooting

Common Issues

Agent not appearing in chat interface

Verify the agent is marked as Active
Check role permissions match your user roles

Chat reaches maximum length unexpectedly

Review the Max Chat Length setting on the agent
Note that only user messages count toward the limit
Start a new chat to continue conversations

Slow response times

Check if Vector DB search is enabled unnecessarily
Review the amount of context being included
Verify the AI Model's performance characteristics

Context not being recognized

Ensure TM AI Context records are active and relevant
Verify context is properly linked to the agent
Check that referenced documents exist and are accessible