Skip to content
Start free

ProductBuilder

Watch the agent build it.Live, against your screens.

The Builder gives every app a live sandbox and an AI agent. Ask for a change and it edits the code, runs lint and typecheck, and saves a version while the preview updates beside it. It starts from your design package, so it builds the screens you confirmed instead of guessing.

Start freeSee pricing
The Builder for a four-app monorepo, session ready. The agent chat shows a request to use the brand logo with an element attached from the preview, answered by Build with Sonnet 5 for $0.20 with lint and type check passed. The live preview has Web, Admin, Marketing, and API tabs, with the Fridgent marketing site open.
Sample project · made in LiteSurface
three agent modes
Build · Plan · Chat
Claude and GPT-6, each at list price
8 models
per request, adjustable up to $10
$2 cap by default
then the sandbox sleeps
3 idle minutes

How it works

Ask, watch,keep what works.

  1. 01

    Open the app

    Start from a concept’s design package, a template, or a prompt. The sandbox wakes with the dev server running.

  2. 02

    Ask the agent

    Describe the change, or select an element in the preview to attach it. Build changes code; Plan and Chat change nothing.

  3. 03

    Watch it work

    The agent reads and edits files and runs lint and typecheck, while the live preview updates in the device frame you pick.

  4. 04

    Keep or restore

    Every turn saves a version with its diff. Restoring one saves a new version, so history is never rewritten.

Costs

Every request,priced and capped.

You choose the model and the effort, and set a spend cap for each request. When a turn finishes, its cost opens into a breakdown by model and token type, at the list prices shown in the picker.

  • Claude Fable 5.1, Opus 5.5, Opus 5, Sonnet 5, and Haiku 4.5; GPT-6 Astra, Sol, and Luna
  • Low, medium, or high effort
  • A spend cap per request: $2 by default, up to $10
  • Workspace budgets and the usage ledger cover all model spend
  • Sandbox time counts against your plan’s build minutes
The Builder model picker: Auto, then Anthropic’s Fable 5.1, Opus 5.5, Opus 5, Sonnet 5, and Haiku 4.5, and OpenAI’s GPT-6 Astra, Sol, and Luna, each with its input and output price per million tokens, and low, medium, and high effort below.
A Builder request’s cost breakdown: $0.20 for 10 model calls on Claude Sonnet 5, split into input from cache, other input, and output tokens, with the per-million list prices.
Sample project · made in LiteSurface

In the product

Your screens,in the agent’s hands.

From the sample project, made in LiteSurface.

  • The Builder’s Design panel showing the imported Fridgent brand from docs/design with its logo and palette colors and links to Mockup Studio and Brand Studio, beside the live marketing site preview that uses the Fridgent logo.

    The Design panel

    Import the latest design package, compare the build with a confirmed screen side by side or as an overlay, and ask the agent to build a screen from it.

  • The Builder home: “What do you want to build?” with a prompt composer, template and framework pickers, example prompts, a row of starter templates with previews, and recent apps.

    Start from a prompt or a template

    Describe the app, pick a framework or let Auto choose, or start from a starter. Apps from a concept arrive with their design package.

Capabilities

Built for real apps,not demos.

  • A live preview

    Device frames for desktop, laptop, tablet, and phone, a tab per app in a monorepo, and a dev server log beside it.

  • Point at what to change

    Select an element in the preview and it is attached to your request, with the file and line it comes from.

  • Files and an editor

    Browse and edit the code yourself. Manual edits save as versions too, automatically or at a save point.

  • Checks after every turn

    Lint and typecheck run after each build turn. The full suite, with tests, acceptance, and a production build, runs on demand.

  • Environment variables

    Stored encrypted and written to the app’s .env.local in the sandbox. The agent sees names, never values.

  • Design package import

    The brand kit, confirmed screens, and specs land in docs/design with icons installed, so the agent builds close to 1:1.

What you getA running app you own.

The code is an ordinary repository you can download or open as a pull request at any time. Solo includes 1 live session and 60 build minutes a month; Team includes 3 sessions and 600 minutes.

  • A live sandbox for every app
  • An agent in Build, Plan, or Chat mode
  • A preview with device frames
  • A version with a diff for every change
  • Every request’s cost, capped by you
  • A repository you can export at any time

From confirmed screensto a running app.

Open the Builder from a concept, a template, or a prompt, and watch the agent build it in front of you.

Start freeTemplates and stacks