app.build by Neon: Spawning 1000s of AI-generated apps at scale with Koyeb
It's 2025, you can build a full-stack app by telling an AI what you want. Inspired by projects and tools like v0, Create, Replit, and Same, Neon built app.build: an open-source AI agent that builds and deploys full-stack web apps, straight from your terminal.
Built for developers, app.build is local-first, extensible, and fully open source. You can tell it what you want to build and it generates code, runs end-to-end tests, deploys the app, and simply gives you a live URL with the app running — all seamlessly deployed on Neon and Koyeb's serverless infrastructure.
In this post, we’ll look at how app.build works under the hood, and how Koyeb helps Neon spawn thousands of applications per day and execute AI-generated code at scale.
What is app.build?
app.build is an open-source AI agent that turns a prompt into a working full-stack web app — right from your terminal. More than just generating code, app.build:
- Creates a GitHub repository with all the code for the app in your account
- Generates fully functional full-stack application code
- Writes and runs end-to-end tests to ensure the app works as intended
- Deploys the app and database using Neon and Koyeb's serverless infrastructure
- Provides a live URL to access the app
Under the hood, it uses a familiar stack (Fastify, Drizzle, React, Vite), pairs it with Claude or Gemini for code generation, and runs on Neon and Koyeb for seamless infrastructure.
Video from Neon
On top of generating web apps, app.build provides more features that you probably need for more complex applications:
- An authentication provider (Neon Auth by default)
- A hosted database (Neon as default)
- Immediate deployment of frontend and backend
- All the code is in how own GitHub account, so you can iterate on it. When you push changes, they are automatically deployed
AI-Generated Apps Need Sandboxed Environments and Scalable Compute Infrastructure
To execute and deploy thousands of full-stack applications per day, Neon needed compute infrastructure that is:
- Fully isolated by default
- Efficient and scales to zero automatically
- Deploys in seconds with integrated CI/CD
To meet those requirements, Neon chose Koyeb, a serverless platform built for running high-performance and scalable AI workloads.
Full Isolation with MicroVMs
Every app built with app.build runs in its own sandboxed microVM on Koyeb. A microVM is a secure, lightweight virtual machine with dedicated compute, memory, and networkin. This ensures:
- No shared runtime or network
- Strict resource isolation between apps
- Secure execution of untrusted AI-generated code
This level of isolation is critical when you're deploying thousands of AI-generated applications with untrusted code per day.
More than compute, Koyeb provides everything needed to get a full-stack app live automatically: automatic domain provisioning, SSL certificates, and native support for Docker and OCI images.
Serverless Infrastructure that Scales Up and Down to Zero
To keep infrastructure efficient and cost-effective at scale, Neon uses scale-to-zero. When an application is idle, it’s automatically paused, meaning the compute resources powering it is scaled to zero. When a request for the app comes in, the application is resumed and spins up in seconds.
Blazing-fast deployments and built-in CI/CD
Once an application is generated, it is deployed in seconds thanks to Koyeb's streamlined deployment pipeline and built-in CI.
When you push updates or iterate on your app, the changes automatically trigger a new deployment. The new version is seamlessly promoted with zero downtime, so users never see a gap or restart.
This fully automated workflow means developers can focus on building, testing, and improving — not managing infrastructure.
Building a Community of Hackers and Vibe Coders
Neon built app.build to solve a hard, exciting problem: turning LLM-generated code into real, working full-stack apps — complete with tests, deployments, and high-performance infrastructure.
By combining agent-based workflows with scalable serverless infrastructure from Koyeb and Neon, app.build closes the gap between AI code generation and functioning software.
This is just the beginning. Today, app.build uses Claude and Gemini to generate code, but it’s designed to be model-agnostic. Soon, you'll be able to bring your own model.
Neon is building app.build as a fully open-source project. You can see all the code, all the commits and all of their planned tasks.
If you’re interested in building real apps from prompts — try app.build, dive into the repo at github.com/appdotbuild, and stay tuned for what comes next!