Skip to main content
The chat interface is where you interact with your agent. You send messages, the agent works on its dedicated computer, and you see the results.

​
Sending messages

Type your request in the input field at the bottom of the screen. The placeholder shows “Ask [agent name]…” to remind you which agent you are talking to.
  • Press Enter to send
  • Press Shift + Enter for a new line
  • Click the arrow button on the right to send
Your message appears on the right side of the chat in a blue bubble.

​
Reading responses

Agent responses appear on the left side. A response typically has two parts:

​
Tool activity

A collapsible summary shows what the agent did on its computer. It looks like:
Searched the web, read 3 files, wrote output.json
Click the summary to expand and see each individual action with its result:
  • Read file / Wrote file — file operations on the agent’s machine
  • Ran command — shell commands the agent executed on its VM
  • Searched the web — web research the agent performed
This is a window into the agent’s work — the same kind of work a human researcher would do on their laptop.

​
Text response

Below the activity summary, the agent’s written response is rendered as formatted text with support for:
  • Markdown formatting, headings, and lists
  • Code blocks with syntax highlighting
  • Tables
  • LaTeX math expressions
  • File links (clickable pill badges that open the file preview)
When the agent references a file it created on its computer, the file appears as a small clickable badge in the response text. Click it to:
  • Preview the file in a side panel (code, PDF, images, markdown)
  • Open a .webapp in a new browser tab as a deployed website

​
Model picker

The dropdown at the bottom-right of the chat input lets you switch AI models. Choose from options like opus 4.6, sonnet 4.6, or gpt 5.4. You can change models between messages — the agent keeps the same computer and workspace regardless of which model powers it.

​
Stopping a response

While the agent is responding, the send button changes to a red stop button. Click it to interrupt the current response. The agent keeps any files it already created on its machine.

​
Shared conversations

In shared conversations, messages from other collaborators appear with their name and avatar. Your own messages are always labeled “You”. Everyone shares the same agent computer and workspace. Viewers in a shared conversation see a read-only input that says “Read-only: viewers can follow shared conversations but cannot send messages.”