AI Agent

The AI agent builds and changes your project for you. Describe what you want in plain language, by text or voice, and the agent writes the pages, the logic and the content, then shows you the result. You choose how much quality to spend on each task with three build levels: Economy, Standard and Ultra.

What the agent can do

• Build from a description - a site, a page, a feature, or a whole app
• Change anything - text, colors, layout, images, logic. Just ask
• Generate images - logos, illustrations and hero images, placed in the page
• Read your whole project - it opens the files it needs before it changes them
• Check its own work - it can open the result in a real browser, look at it, and fix what it finds
• Answer questions - about your project, or about what to do next

How to use it

1. Open a project. The chat is the panel beside your preview
2. Describe what you want, in any language. You can type or dictate
3. Watch it work, then review the preview and ask for changes

New accounts start with trial credits and no credit card, so you can build before you pay. On the free plan the agent runs on the Economy level; paid plans unlock Standard and Ultra. See Pricing for current limits.

Build quality: Economy, Standard, Ultra

One setting decides how much the agent spends on a task. Open it from the build-quality control in the chat toolbar (it is titled Choose build quality), and switch levels whenever you like - per chat, or as your default in Settings → You → AI defaults.

Economy

Good for simple edits, fixes, and basic pages. It is the cheapest level and the default for a new account. On complex tasks it can repeat itself or get stuck; if that happens, press Stop and switch up rather than letting it retry.

Standard

Beautiful, detailed design, the cleanest results and the fewest errors. This is the level most paid work should sit on, and it becomes your default when you subscribe.

Ultra

Our most capable model, for the hardest, highest-stakes builds. It is also the most expensive level and can use credits several times faster than Economy or Standard, so choose it on purpose for a specific task, then switch back.

Models

Each level comes with a default model, and you can pick a different one for that level from the same build-quality panel. The list is curated: we keep the current frontier models from several providers and retire the rest, so the exact line-up changes as new models ship.

Economy models

• Gemini 3 Flash (default) - Google's fast model
• GLM 5.3
• DeepSeek V4 Pro
• DeepSeek V4.1 Flash
• GPT-5.6 Luna

Standard models

• Claude Opus 5 (default)
• GPT-5.6 Sol
• Claude Sonnet 5
• Grok 4.6
• Gemini 3.8 Flash
• Claude Opus 4.6
• Kimi K3

Ultra models

• Claude Fable 5.1 (default) - the most capable model we offer
• GPT-6 Astra

When a new frontier model ships it usually appears here within about a day. Older models are retired from the picker, but a project you already built keeps working.

What a task costs

Every model shows a cost meter next to its name, from lowest to highest, so you can see what you are choosing before you send the message. The actual charge depends on the real work done: the model, the size of your project and the length of the conversation. Many small messages cost less than one credit.

A practical habit: keep Economy for small edits, move to Standard for real design work, and reach for Ultra only on the hard parts. See How to save credits and How credits work.

What to ask

Building

Start with what you want to achieve:

• "Build a landing page for my photography business"
• "Create a pricing page with three plans"
• "Add a contact form with email, phone, and message fields"

Changing things

Tell the agent what to fix or update:

• "Make the header background dark blue"
• "Move the testimonials section above the pricing"
• "Replace the hero image with something more professional"

Getting help

Ask questions about your project:

• "How do I add a Google Maps embed?"
• "What would look better here - a grid or a carousel?"
• "Can you suggest a better color scheme?"

While it works

Live preview

The preview updates as the agent works, so you see the result taking shape.

Checkpoints

A change you do not like is not a problem. Playcode saves checkpoints as you go, and restoring one puts your code, files and database back to that point together. See Checkpoints.

Chat history

Conversations are saved. Rename or delete them, and pick up an old one where you left off.

Visual editing

Click an element in the preview and say what to change. You do not have to describe where it is on the page.

Tips for better results

• Match the level to the task - Economy for small edits, Standard for design work, Ultra for the hard parts
• Describe the outcome - "a modern landing page for a bakery" works better than "make it nice"
• Let it ask - if something is unclear the agent will ask rather than guess
• Build in steps - get the basics right, then add one feature at a time
• Say what you think - tell it what you like and what to change
• Stop a run that is going wrong - pressing Stop early saves credits you would spend on a retry