Agent Mode

Agent Mode is an autonomous multi-step app builder. Instead of generating code from a single prompt, the AI breaks your request into a series of tasks and executes them sequentially — planning, building, testing, and fixing errors automatically.

The AI analyzes your prompt and creates a structured task plan before writing any code. Tasks are executed one by one, with each step building on the results of the previous one. Errors are detected and auto-healed between steps, so no manual intervention is required. The final result is a complete, working app that matches your original prompt.

Agent Mode is best suited for complex apps that would be difficult to build in a single generation pass. Use it when your prompt involves multiple features, screens, or interconnected components.

• ●Complex multi-feature apps — e.g., "Build a project management tool with kanban boards, team members, and a dashboard."
• ●Apps with many screens — e.g., "Create an e-commerce site with product listing, cart, checkout, and order history."
• ●Apps that need multiple iterations — Agent Mode refines as it builds, catching and fixing issues along the way.

Agent Mode follows a structured pipeline from prompt analysis to final delivery. The AI handles every stage autonomously so you can sit back and watch your app come together.

Agent Mode provides real-time visibility into what the AI is doing. You can monitor each step and intervene if needed.

• ●The task plan is displayed in the UI before execution begins — you can see what the AI intends to build.
• ●Each task shows its status in real time: pending, in progress, completed, or failed.
• ●You can watch code being generated and modified as each task executes.
• ●Cancel at any time — the app preserves whatever has been built up to the cancellation point.

Agent Mode sessions have limits to prevent runaway execution and ensure system stability. These limits are generous enough for most apps but worth knowing about for very large projects.

• ●20 tasks per session — if your request requires more, break it into multiple sessions.
• ●12-minute timeout per task — individual tasks that take longer are terminated.
• ●1-hour inactivity timeout — the session ends if no task is executed for 60 minutes.

Both features build and modify apps, but they serve different purposes and work differently. Understanding when to use each one will help you work more effectively.

• ●Agent Mode builds autonomously from scratch — you provide the prompt and the AI handles everything.
• ●Chat Refinement iterates on existing code — you guide the AI with specific change requests.
• ●Agent Mode is best for initial app creation with complex requirements.
• ●Chat Refinement is best for making targeted changes to an existing app ("move the sidebar to the right", "add a search bar").

FAQs