Skip to main content
The Chat page is where you interact with your agent. It’s a real-time conversation interface that shows your messages, agent responses, and any tools the agent uses along the way. Open your agent → Chat to start talking.

Connection Status

At the top you’ll see “CONNECTED” with a green dot when your agent is ready. If the connection drops, the status updates and you can refresh to reconnect.

Switching Models

Use the dropdown in the top-right to switch between any models you’ve enabled for this agent. You can switch mid-conversation — useful for:
  • Using a faster model for quick questions
  • Switching to a more capable model for complex reasoning
  • Comparing how different models handle the same prompt
The dropdown shows models grouped by provider (e.g., anthropic/claude-sonnet-4-6).

Display Settings

Click the settings icon (sliders) to toggle what’s visible in the chat:
SettingWhat it does
MarkdownRender markdown formatting in responses
ThinkingShow the agent’s extended thinking process
Tool CallsShow tool use blocks (commands, API calls, etc.)
Image PreviewDisplay images inline
TimestampsShow when each message was sent
AvatarsShow profile pictures next to messages
These are personal preferences — they don’t affect how your agent works, just how you see the conversation.

Tool Calls

When your agent uses a tool, you’ll see it in the chat as an expandable block:
  • Tool name — What tool was called (e.g., cron, bash, web_search)
  • Arguments — What parameters were passed
  • Result — What came back
Click “Show more” to expand the full details. This is helpful for debugging or understanding exactly what your agent did.

Fullscreen Mode

Click the expand icon in the top-right to go fullscreen. This hides the sidebar and gives you more room for the conversation. Click again to exit.

Tips

  • Your agent remembers the full conversation context, so you can refer back to earlier messages
  • If the agent seems stuck, try rephrasing or breaking your request into smaller steps
  • Tool calls can take a few seconds — the interface shows when the agent is working