Chat Guide

Chat with agents using natural language to perform various tasks. This guide explains effective conversation methods and chat features.

Start Chatting

1. Access Chat Page

Click the Chat button in the left sidebar.

2. Select Agent

Activated agents are displayed as cards.

Card Info	Description
Agent Name	Agent identification name
Description	Agent's role or function
Connected Resources	MCP server, data source icons

3. Start Conversation

1. Click the agent card you want to use
2. When the drawer opens, review the agent details
3. Click the Start Chat button
4. Navigate to the chat screen

Chat Interface

Screen Layout

┌─────────────────────────────────────┐
│ Agent Name              [Settings][X]│
├─────────────────────────────────────┤
│                                     │
│         Conversation Area           │
│                                     │
│  [User] Hello                       │
│                                     │
│  [Agent] Hello! How can I help you? │
│                                     │
├─────────────────────────────────────┤
│ [Message Input]              [Send] │
└─────────────────────────────────────┘

Send Message

1. Enter your message in the input field at the bottom
2. Press Enter key or click the Send button
3. The agent generates a response

Line Break

Use Shift + Enter to add line breaks within a message.

Effective Conversation Tips

Ask Clear Questions

❌ Bad: "How do I do this?"
✅ Good: "Summarize the access control requirements from the NIST 800-53 document"

Provide Context

❌ Bad: "Summarize it"
✅ Good: "Summarize the security policies mentioned in our previous conversation in table format"

Break Down Complex Requests

Divide complex tasks into multiple steps:

1. "First, extract the key terms from the document"
2. "Categorize the extracted terms"
3. "Write a summary for each category"

Tool Calls (MCP)

Agents connected to MCP servers can call external tools.

Tool Call Example

User: "What's the weather in Seoul today?"

Agent: [Calling Web Search tool...]
       ────────────────────────
       Tool: weather_search
       Input: {"location": "Seoul"}
       Result: {"temp": 25, "condition": "sunny"}
       ────────────────────────

       Today's weather in Seoul is sunny with a temperature of 25°C.

Tool Call Display

When a tool is called, the following information is displayed:

• Called tool name
• Input parameters
• Returned result

Session Management

Start New Session

Click the New Chat button on the chat screen to start a new conversation session.

Session History

View previous conversation sessions and continue conversations.

1. Open the Session List panel on the left side of the chat screen
2. Click the session you want to continue
3. Previous conversation content loads

Delete Session

1. Hover over the session to delete in the session list
2. Click the Delete icon
3. Click Confirm in the confirmation popup

Caution

Deleted sessions cannot be recovered.

Response Format

Markdown Support

Agent responses support markdown format:

• Bold text, italics
• Lists (ordered/unordered)
• Code blocks
• Tables
• Links

Code Blocks

Responses containing code provide syntax highlighting and copy functionality.

# Example code generated by agent
def hello():
    print("Hello, World!")

Troubleshooting

Slow Response

• Complex questions take longer to respond
• MCP tool calls add external service response time
• Large data source searches may take time

Inaccurate Response

• Try rephrasing your question more specifically
• Check if relevant data sources are properly connected
• Check if required MCP servers are activated

Next Steps

• DataSource Management - Add knowledge to agents
• MCP Installation - Connect external tools
• Troubleshooting - Solve common problems