The Chat interface is where you interact with AI Agents. Access it from the Chat page in the main navigation.
Starting a Conversation
From the Chat landing page you can:
- Start a new conversation by typing a message in the input field
- Resume an existing conversation by clicking on a ticket in the list
- Filter tickets using the "My Tickets" toggle to show only your conversations
- Search tickets by title or content (appears when you have more than 5 tickets)
Each conversation is called a session (or "ticket"). Sessions persist so you can return to them at any time.
The Chat View
The chat view has three areas:
Session List (left sidebar)
A list of all your chat sessions with:
- Session title (auto-generated from the first message)
- Active run indicator (green dot when the agent is working)
- Source icon — for web, for Slack
Message Area (center)
Shows the conversation history:
- Your messages — displayed with your avatar
- Agent responses — text with markdown formatting, code blocks, tables, and links
- Tool calls — visual cards showing what the agent is doing (see below)
- Thinking sections — expandable blocks showing the agent's reasoning process
Input Area (bottom)
Type your message and press Enter to send. The agent processes your request and streams the response in real time.
Tool Call Rendering
When the agent uses a tool, you see a visual card in the chat:
SQL Query
Interactive table with CSV download
Chart
Interactive Plotly with zoom & hover
Email / Slack
Formatted preview with send status
Ask Question
Multiple choice buttons or text input
Session Status
Active — normal conversation
Waiting — agent asked a question
Closed — conversation finished
Session Management
- Delete a session — click the delete button with confirmation
- Edit messages — modify your sent messages to re-run the agent's response
- Share messages — share individual messages with colleagues
Real-Time Streaming
⚡
Agent responses stream in real time using Server-Sent Events (SSE). You see text appearing word by word, tool calls executing, and results rendering as they complete. If you navigate away during an active run, the agent continues working in the background — results will be visible when you return.