Skip to main content
Agent mode is a full-featured coding agent powered by frontier models — the same code quality as Cursor and Claude Code. It directly edits your prototype with access to standard coding tools like grep, terminal, file operations, and code generation.

​
How Agent Mode Works

Agent mode works like a standard AI coding assistant. You describe what you need, and it edits the code directly in your current prototype (frame).

​
Standard Coding Tools

Agent mode has access to all the tools you’d expect from a modern coding agent:
  • Code generation — Write new code, components, and features
  • File operations — Create, edit, delete, and move files
  • Grep/search — Search through your codebase to understand context
  • Terminal — Run commands, install packages, execute scripts
  • Code analysis — Understand existing code patterns and structure

​
Powered by Frontier Models

Agent mode uses the latest frontier AI models — the same level of code quality and understanding as Cursor and Claude Code. It writes production-ready code that fits your patterns and style.

​
When to Use Agent Mode

Quick Fixes

Fix bugs, update styles, or make small changes

Refactoring

Improve code structure, extract components, or optimize

Feature Updates

Add functionality to existing features

Code Improvements

Enhance existing code based on feedback

​
Example Workflow

  1. You request: “Fix the button spacing in the header component”
  2. AI analyzes: Uses grep to find the header component, understands the current styling
  3. AI edits: Updates the spacing directly in the code
  4. You see the change: The prototype updates with the fix applied
Or:
  1. You request: “Add error handling to the API call in the user profile”
  2. AI searches: Finds the API call, understands the context
  3. AI writes: Adds try/catch blocks and error states
  4. Done: Error handling is now in place
Agent mode is perfect for tasks you’d normally do in Cursor or Claude Code. For creating multiple variations at once, use Prototype mode instead.

​
Working with Prototypes

Each prototype (frame) can have multiple chats in Agent mode, with separate contexts per conversation. For details on how chats relate to prototypes and how to manage multiple chats, see Chat Modes & Commands.

​
Best Practices

Be specific — The more context you provide, the better the results. Reference specific files, components, or patterns.
Use @Codebase — When working with a connected repository, mention files using @Codebase to give the AI better context about your codebase structure.
Iterate incrementally — Make small, focused changes rather than large refactors. It’s easier to review and understand.