Weblab: AI Website Builder for React and Next.js Codebases

Weblab is an AI website builder for teams that already have a React or Next.js codebase. The AI is constrained to your real components and design tokens. Every change ships as a pull request reviewable in GitHub. Bring your own model — Claude, GPT, Gemini, DeepSeek, Mistral, or local Ollama. Open source and self-hostable.

AI Website Builder

AI website builder for the codebase you already own

Most AI website builders regenerate a new app from a prompt. Weblab edits your real React components, respects your design tokens, and writes every change as a pull request to your GitHub repo.

Constrained AI, not generative chaos

Generative AI website builders sound magical and ship slop. The model invents a new button, a new spacing scale, a new color, a new component name. You end up with a tree of orphan code that ignores everything your team already built.

Weblab reverses that contract. The AI sees your existing components, your tokens, your conventions. It is graded on whether the change fits your system — not on whether it generates the fanciest thing.

The output is a git diff. Your engineers review it. Bad diffs get rejected like any other PR. Good ones merge and ship.

What the AI does

Design system aware edits

AI knows your tokens and your component library. Asks for the "primary button" and gets your `<Button variant="primary">` — not a fresh `<button>` styled from memory.

Page-level composition

Generate a new page or section from a prompt. The AI assembles it from components that already exist in your repo, wires up routes, and writes the file in the correct App Router location.

Refactors against real code

Rename props, extract a component, change a token. The AI rewrites every call site and ships a single mergeable PR.

Visual edits, code source-of-truth

Designers move things on the canvas. Engineers see the diff. The two stay in lockstep because both views read the same JSX.

Multi-model, your choice

Claude, GPT, Gemini, DeepSeek, Mistral, or local Ollama. Switch per project. Bring your own API key in self-host mode.

Reasoning effort dial

Fast for "tighten this spacing", deep for "rewire auth across five routes". Pay only for the depth you need.

How it compares to other AI builders

CapabilityWeblabGenerative AI builders
Starting pointYour existing React/Next.js repoA blank canvas + a prompt
OutputPR to your GitHub repoNew hosted app on the vendor
Design systemAI is constrained to your tokens and componentsAI invents components and styles each time
Code ownershipLives in your repo from day oneLives on the vendor — export is an escape hatch
Model choiceClaude, GPT, Gemini, DeepSeek, Mistral, local OllamaUsually one model, vendor-chosen
Engineer review surfaceGitHub PR with CI, comments, file-by-fileGenerated chat thread + preview
Self-hostYes — open sourceNo

An AI assistant that builds and edits your website inside the canvas. In Weblab the AI is constrained to your real React components, your design tokens, and your existing routes. It doesn't generate a new app from scratch — it modifies the codebase you already own.

Those tools generate a new app from a prompt. The output is a fresh project on their platform. Weblab works on your existing repo. The AI reads your components and writes diffs against them as pull requests — your engineers review the change in GitHub like any other PR.

Claude Opus, Sonnet, Haiku; GPT-5.5; Gemini Pro; DeepSeek; Mistral Codestral; or your local Ollama. You can also bring your own API key. Reasoning effort is tunable from fast to deep per request.

Not by default. It is constrained to the components and tokens already in your design system. If you explicitly ask for a new one, it scaffolds it in the same style as your existing components and writes the matching token entries.

It can write component tests, snapshot tests, and visual regression specs alongside the change. For schema migrations, it proposes the migration file and the corresponding code change in a single PR, but you review and run the migration yourself.

In the hosted cloud version, the AI runs in our infrastructure with the provider you choose. In the self-hosted version, the AI talks directly to the provider's API using your key — nothing routes through us.

Free tier with daily message limits. Paid Pro tiers from $25/month for higher limits, with eleven pricing tiers up to enterprise. Self-hosted is free with your own API key.

Yes — every AI change is a git diff. You see the PR in GitHub, with proper file-by-file review, comments, CI checks, and merge controls. If the AI got it wrong, you reject the PR like any other.

Constrained AI.
Reviewable diffs.