The Meta Agent chat interface provides a full-featured workspace for interacting with the Kubiya platform through natural language.
Compose Workspace (/chat)
Click Compose in the sidebar for the full-screen workspace:
Features include:
- Conversation sidebar with history and search
- Full-width message display for extended conversations
- Insights panel with AI-powered suggestions
- Multiple input controls for connectors, models, and execution modes
Floating Panel (Cmd+J)
Press Cmd+J (Mac) or Ctrl+J (Windows/Linux) from any page for quick access without leaving your current context.
| Use Panel | Use Workspace |
|---|
| Quick questions | Extended conversations |
| While on other pages | Dedicated focus |
| Single queries | Multi-turn dialogues |
At the bottom of the chat, you’ll find several control buttons:
| Control | Description |
|---|
| Add Files | Upload photos or files to include in your conversation context |
| Attach Repositories | Select GitHub repositories to attach to the session (up to 10) |
| Remote Task Queues | Select a worker queue for distributed task execution |
| Connectors | Enable integrations like GitHub, Sandbox, Slack |
| Model Selector | Choose the AI model for this conversation |
Click the connectors button to access external integrations:
Available connectors include:
- GitHub - Repository access, PRs, issues
- Sandbox - Safe code execution environment
- Slack - Team messaging and notifications
- AWS - Cloud resource management
- Jira - Issue and project tracking
- Create Apps - Generate and deploy applications
Enable multiple connectors simultaneously. The Meta Agent will automatically use the appropriate connector based on your request.
Model Selection
Click the model selector to choose your preferred AI model:
Available models include:
- Claude Sonnet 4.5 - Balanced performance and capability (default)
- Claude Opus 4.5 - Maximum capability for complex reasoning
- Claude Sonnet 4 - Fast, efficient for straightforward tasks
- Additional models from OpenAI, Amazon Nova, Qwen, and more
Each model shows input/output pricing for cost awareness.
Act vs Plan Mode
Toggle between execution modes using the radio buttons above the input:
| Mode | Behavior |
|---|
| Act | Execute tasks immediately with real actions |
| Plan | Generate plans without execution for review first |
Use Plan Mode when you want to:
- Review what actions will be taken before execution
- Understand the steps involved in complex operations
- Get clarification on approach before committing
Remote Task Queues
Click the queue selector to delegate tasks to remote workers:
- View all available task queues (e.g., Dev, Production, GPU workers)
- See worker capacity and availability for each queue
- Select a queue to route execution to specific infrastructure
- Ideal for tasks requiring access to private resources or specialized hardware
Conversation Management
The conversation sidebar provides:
- Search: Find past conversations by keyword
- History: Browse conversations grouped by date (Today, Yesterday, This Week)
- Message count: See how many exchanges in each conversation
- Model info: Know which model was used for each conversation
Session Controls
- New Conversation: Start fresh with
New Conversation button
- Stop Button: Halt streaming responses or cancel long-running operations
- Clear Chat: Reset the conversation context
Insights Panel
The Meta Agent proactively analyzes your infrastructure and surfaces intelligent suggestions:
- Security Alerts - Detected vulnerabilities requiring attention
- Health Issues - Unhealthy resources or degraded services
- Optimization Opportunities - Cost savings or performance improvements
- Pattern Insights - Trends from cognitive memory analysis
Click Customize suggestions to personalize what insights you see.
Keyboard Shortcuts
| Shortcut | Action |
|---|
Cmd/Ctrl + J | Open/close Meta Agent panel |
Enter | Send message |
Shift + Enter | New line in message |
Escape | Close Meta Agent panel |
Cmd/Ctrl + B | Toggle sidebar |
What’s Next