AI platform to build full-stack Web3 apps from prompts, in-browser with Flow and graph intelligence.
This project is a next-generation, in-browser AI-powered Web3 development platform that empowers anyone—regardless of technical background—to create, customize, and deploy decentralized applications with integrated smart contracts using natural language prompts.
The platform removes the complexity of setting up environments, learning blockchain syntax, or dealing with backend infrastructure. Instead, users simply describe what they want to build, and the system—powered by Anthropic Claude AI—translates that intent into fully functional Web3 applications, including smart contracts, frontend UIs, and wallet integrations.
At its core, the platform leverages the Flow blockchain for its user-friendly architecture and scalable smart contract layer. Users can connect wallets, deploy contracts, and test dApps—all from within the browser.
What truly sets this project apart is its ability to run an entire development environment inside the web browser using StackBlitz’s WebContainer technology. This allows code to execute instantly without server infrastructure or external deployments. Developers (and non-developers) can see live previews, edit smart contracts, and interact with blockchain logic in real time—no installations needed.
To make the experience even more dynamic, the system integrates The Graph and HyperGraph for indexing smart contract events and metadata. This means that any deployed application can be queried and explored visually, making the entire dApp lifecycle—from creation to data visibility—transparent and interactive.
We also utilize IndexedDB in the browser to save user data, application state, and custom project files locally. This ensures session persistence, offline access, and quick recovery without user frustration.
We built this project by combining a stack of modern Web3 and Web2 technologies to enable a fully browser-based development environment for smart contract and dApp creation. The goal was to eliminate technical barriers for both developers and non-developers, allowing them to build and deploy blockchain applications with natural language.
At the core, we used the Flow blockchain for its developer-friendly smart contract model (Cadence), paired with Flow wallets to manage user accounts and on-chain interactions. For smart contract deployment and wallet connection, we integrated Flow’s JS SDK into our frontend.
To handle natural language prompts, we used Anthropic Claude AI. Claude acts as an LLM agent that takes user instructions and converts them into application logic, smart contract code, or configuration scripts. These LLM calls are triggered from the frontend using API calls, and Claude’s output dynamically updates the coding environment.
One of the key innovations in this project is the use of StackBlitz web containers. This allowed us to run a real, fully isolated development environment in the browser itself, without needing a backend server. Users can write, edit, and test their blockchain applications live—entirely client-side. This setup is powered by WebAssembly under the hood and creates a seamless local-like development experience right in the browser.
For data indexing and smart contract metadata, we used The Graph. We configured subgraphs to collect contract events and structured data, which we then queried via HyperGraph for live updates within the application. This enabled features like displaying real-time contract metadata and historical transaction data in the frontend UI.
We also used IndexedDB to persist project data, such as code, settings, and recent activity, directly in the browser. This helped us offer offline support and ensure a smoother experience when users switch between sessions or go offline.
A particularly hacky and clever part of the build was wiring Claude’s LLM output directly into a containerized file system inside the browser. We had to manage sandboxing limitations and emulate a file-based coding experience using virtual FS APIs. This allowed us to turn LLM-generated code into live-running dApps instantly.