Open Lovable

Open Lovable lets you chat with an AI to build React apps, with a specific, differentiating capability: it can scrape an existing website (via Firecrawl) and recreate it as a working React app, including extracting brand styles and screenshots to match the original’s look. Generated apps run in a live sandbox with a Vite dev server, streamed AI code application, and real-time error/log monitoring so issues surface immediately rather than after a full rebuild.

Built by the Firecrawl team as an open, self-hostable example of the “AI app builder” category popularized by Lovable.dev, v0, and Bolt.new, it requires a Firecrawl API key (for scraping) plus at least one LLM provider key (Gemini, Anthropic, OpenAI, or Groq) to run — it’s not a zero-dependency tool, but every dependency is a standard API key rather than a proprietary backend.

The project is MIT licensed and explicitly points to Lovable.dev as “a complete cloud solution” for users who want a hosted version instead of self-hosting.

What You Get

• AI-driven React app generation from a chat interface, running in a live Vite sandbox
• Website-to-React cloning: scrape an existing site and recreate it as working code, including brand styles
• Streamed AI code application with real-time Vite error and log monitoring during generation
• Support for multiple LLM providers (Gemini, Anthropic, OpenAI, Groq) instead of being locked to one

Common Use Cases

• Quickly cloning the structure and styling of an existing website into an editable React codebase
• Prototyping a React app through conversational AI instead of hand-writing boilerplate
• Self-hosting an AI app-builder workflow instead of depending on a hosted service like Lovable.dev or v0
• Extracting a site’s brand styles (colors, fonts) automatically as a starting point for a new build

Under The Hood

Architecture Open Lovable is a Next.js app whose real logic lives in its API routes: scrape-website and scrape-screenshot pull source material from a live URL via Firecrawl, extract-brand-styles derives a design system from it, create-ai-sandbox spins up an isolated Vite environment, and apply-ai-code-stream streams generated code into that sandbox while monitor-vite-logs/report-vite-error surface build issues back to the user in real time. This pipeline — scrape, extract, generate, sandbox, monitor — is what turns a website URL into working, editable code rather than a static screenshot clone.

Tech Stack Next.js and TypeScript for the app itself, Firecrawl for web scraping (a required dependency, not optional), a Vite-based sandbox for running generated code live, and pluggable LLM support across Gemini, Anthropic, OpenAI, and Groq.

Code Quality The API route structure is granular and single-purpose (one route per pipeline step) rather than one large generation endpoint, which makes the scrape→generate→sandbox flow easier to reason about; GitHub activity shows relatively low recent commit velocity despite the project’s popularity, worth noting if you need active upstream fixes.

What Makes It Unique Most AI app builders start from a blank prompt; Open Lovable’s distinguishing feature is starting from an existing website URL, scraping and extracting its actual brand styles, then regenerating it as React code in a live, monitored sandbox — closer to a website cloning pipeline than a generic chat-to-code tool.

Self-Hosting

Licensing Model MIT licensed — fully open source and self-hostable with no license key.

Self-Hosting Restrictions None in the code itself, but running it requires a Firecrawl API key (for the scraping functionality) and at least one LLM provider API key — these are external service dependencies, not license gates in the Open Lovable code.

License Key Required No license key for the software itself; API keys for Firecrawl and an LLM provider are required to run it.