Basic Usage

This guide covers the Playground interface and how to interact with AI agents effectively.

Interface Layout

Playground consists of two main panels:

• Left Panel: Live desktop stream from the sandbox
• Right Panel: Chat interface for AI communication

The top bar provides model selection and session controls.

Desktop Panel

Live Stream

The desktop panel shows real-time video of the sandbox. You can see:

• Mouse movements and clicks
• Keyboard input
• Application windows and interactions
• Visual feedback for AI actions

Read-Only View

The desktop is controlled by the AI agent. Direct interaction (clicking or typing) is not available. To guide the AI, use the chat interface.

This ensures:

• Consistent AI behavior
• Proper action tracking
• No conflicts with AI operations

Chat Panel

The chat panel is where you communicate with the AI agent.

Sending Messages

Type your request in the input field at the bottom and press Enter to send.

For multi-line messages, use Shift+Enter to add line breaks without sending.

Message Display

Your messages and AI responses appear in the chat history. The AI may show:

• Thinking process (if the model supports it)
• Action progress updates
• Completion status
• Error messages if something fails

Model Selection

The model dropdown at the top lets you choose which AI model to use. Each model has different capabilities for UI understanding, visual processing, and speed.

Changing the model only affects new messages. Previous messages retain their original model.

See Available Models for details on each option.

Session Controls

New Chat

Start a new conversation while keeping the same sandbox running. This clears the chat history and context, useful when switching to an unrelated task.

The sandbox continues running, so any applications or files remain accessible.

System Prompt

Advanced users can customize the AI's behavior by modifying the system prompt. Access this through the settings in the top bar.

Changes affect new messages in the current session.

Writing Effective Prompts

Be Specific

Provide clear, detailed instructions:

Open Chrome and navigate to github.com/lybic

Instead of:

Open the browser

Include Context

Specify application names, file names, and URLs explicitly:

Create a file named "notes.txt" with the text "Meeting summary"

Break Down Complex Tasks

For multi-step operations, either describe all steps or send them one at a time:

1. Open a terminal
2. Run "python --version"
3. Show the output

Monitoring Progress

The AI will show its progress in the chat. Watch both the chat and desktop stream to:

• Verify actions are executing correctly
• Catch errors early
• Understand the AI's approach

If something goes wrong, the AI will usually explain the error in chat.

Sandbox Applications

The sandbox comes with pre-installed applications:

• Chrome: Web browser
• Terminal: Command-line access
• Text editors: Various editors available
• Development tools: Basic programming tools

Specify the application name in your requests for better results:

Open Chrome and go to lybic.cn

Open a terminal and run "ls -la"

Session Management

Sandbox Lifecycle

Sandboxes run according to your organization's plan configuration:

• Automatically stopped after idle timeout
• Maximum lifetime based on plan
• Sandbox hours tracked for billing

Resource Usage

Each active session consumes:

• Sandbox Hours: Time the sandbox is running
• Agent Credits: AI model operations

View usage in your Dashboard billing section.

Common Issues

Slow Response

If the AI or video stream is slow:

• Check your internet connection
• Reduce browser tab count
• Try a simpler task

Connection Lost

If the session disconnects:

• Refresh the page
• Check your organization's resource availability
• Contact support if the issue persists

Next Steps