Chat Interface
Chat Interface
Using the Agent Chat window
The Agent Chat window is a native Slate-based UI for conversing with AI agents directly in Unreal Editor.
Opening the Chat Window
- Menu: Tools > Agent Chat
- Shortcut: Configure in Editor Preferences
The window is dockable—drag it to any panel location in your editor layout.
Interface Elements
Header Bar
| Element | Description |
|---|---|
| Agent Dropdown | Select which agent to use |
| Model Dropdown | Choose model (if agent supports multiple) |
| Mode Dropdown | Select operational mode |
| Connect/Disconnect | Toggle agent connection |
Message Area
Displays the conversation with different message types:
- User messages: Your prompts (right-aligned)
- Assistant messages: AI responses (left-aligned)
- Tool calls: Expandable panels showing tool invocations
- Thought blocks: Collapsible AI reasoning (when available)
Input Area
| Element | Description |
|---|---|
| Text input | Multi-line message composition |
| Send button | Submit message (or press Enter) |
| Attachment indicator | Shows attached context |
| Clear button | Reset conversation |
Message Types
Text Messages
Standard conversation messages with markdown support:
- Bold, italic,
code - Code blocks with syntax highlighting
- Lists and headers
- Links
Tool Calls
When the AI uses a tool, you'll see:
- Tool name and description
- Input parameters (expandable)
- Execution status
- Result (expandable)
Click to expand/collapse tool details.
Thought Blocks
Some agents (like Claude) show their reasoning:
- Collapsible by default
- Shows step-by-step thinking
- Useful for understanding complex decisions
Permission Requests
When a tool needs approval:
- Dialog appears with action details
- Allow: Proceed with the action
- Deny: Cancel the action
- Always Allow: Don't ask again for this tool
Keyboard Shortcuts
| Shortcut | Action |
|---|---|
| Enter | Send message |
| Shift+Enter | New line in message |
| Escape | Clear current input |
| Ctrl+L | Clear conversation |
Streaming
Responses stream in real-time:
- Text appears word-by-word
- Tool calls show as they're invoked
- Progress visible during long operations
Tips
- Be specific: Clear prompts get better results
- Provide context: Attach relevant nodes or assets
- Review tool calls: Expand to see what the AI is doing
- Use permissions: Allow trusted operations, deny uncertain ones
- Start fresh: Clear conversation if context becomes confused