Compose Workspace

Documentation Index

Fetch the complete documentation index at: https://docs.kubiya.ai/llms.txt

Use this file to discover all available pages before exploring further.

The Compose Workspace is your primary interface for extended conversations with the Meta Agent. Unlike the floating panel, the workspace provides a dedicated full-screen environment with conversation history and enhanced controls. The workspace features a three-panel layout:

• Left sidebar: Conversation history with search and new conversation button
• Center panel: Active conversation with messages and responses
• Input bar: Controls for attachments, model selection, and execution mode

​

Access

Click Compose in the sidebar to navigate to the workspace at /chat.

​

Workspace Layout

​

Conversation Sidebar

The left panel manages your conversation history:

• New Conversation - Start a fresh conversation
• Search - Find conversations by content
• Conversation List - Browse previous conversations grouped by time:
  • Today
  • Yesterday
  • Last 7 days
  • Older

Each conversation shows:

• Title (auto-generated from first message or renamed)
• Model used (e.g., Claude Sonnet 4.5)
• Message count
• Timestamp
• GitHub/repository attachments (if any)

Right-click on any conversation to:

• Rename the conversation
• Delete the conversation

​

Main Chat Area

The center panel displays the current conversation: Header:

• Conversation title
• Jobs indicator (shows active background tasks)
• Share button
• Settings menu

Messages:

• User messages appear on the right
• Agent responses appear on the left with:
  • Reasoning sections (expandable)
  • Tool calls with results
  • Visual/JSON toggle for structured data
  • Copy button for responses

​

Input Bar

The bottom input area provides powerful controls:

​

Input Modes

Act Mode (Default):

• Execute tasks immediately
• Real actions are performed
• Results are returned in real-time

Plan Mode:

• Generate execution plans without taking action
• Review steps before committing
• Ideal for complex or sensitive operations

​

Conversation Management

​

Starting a Conversation

1. Click + New Conversation in the sidebar
2. Type your request in the input bar
3. Press Enter or click the send button

​

Continuing Conversations

• Click any conversation in the sidebar to resume
• All context from previous messages is preserved
• The Meta Agent remembers what you discussed

​

Attaching Context

Files: Click the + button to upload:

• Images for visual analysis
• Documents for reference
• Data files for processing

Repositories: Click the code icon to:

• Search connected GitHub repositories
• Select up to 10 repositories
• Provide code context for the agent

Remote Execution: Click the server icon to:

• View available task queues
• See worker availability
• Route tasks to specific infrastructure

​

Working with Responses

​

Tool Calls

When the Meta Agent uses platform tools, you’ll see:

• Tool name and status
• Execution time
• Expandable results section
• Visual or JSON view toggle

​

Structured Output

For data responses, toggle between:

• Visual - Formatted tables and cards
• JSON - Raw data for copying

​

Reasoning

Expand the “Reasoning” section to see how the agent:

• Interpreted your request
• Selected tools to use
• Arrived at conclusions

​

Header Controls

​

Keyboard Shortcuts

​

Best Practices

1. Use descriptive first messages - The first message becomes the conversation title
2. Attach relevant repositories - Code context improves suggestions
3. Review tool calls - Understand what actions are being performed
4. Use Plan mode for sensitive operations - Review before executing
5. Organize with projects - Filter conversations by project scope

​

Related Pages

• Meta Agent - Learn about Meta Agent capabilities
• Meta Agent Panel - Quick access from any page
• Task Kanban - Track executions created from conversations