From AI-Built App to Production in 2026: Where to Run What You Made with Cursor, Claude Code or Lovable

For the fastest live URL, publish from the builder itself (Lovable, Bolt or Replit) or push the repository to a hosted platform such as Vercel, Render, Railway or Fly.io. To run the app on servers you control, pick a self-hosted PaaS (Coolify, Dokploy, Kamal) for one small server, a BYOC platform (Northflank, Qovery) if your company already runs on AWS, Google Cloud or Azure, or Ankra to turn a GitHub repository into a live URL in your own cloud account, including European providers like Hetzner, OVHcloud and UpCloud, with a generated pipeline that tests, scans, previews and rolls back. Below, what production adds, then each route compared.

What production adds to a demo URL

A demo URL proves the code runs once. Production keeps it running when people depend on it, and lets you change it safely. You do not need to be an infrastructure engineer to check for these six things.

• A database that is backed up. User data lives in a real database, and a copy goes on a schedule to a place you can restore from.
• Secrets kept out of the code. API keys and passwords are settings the platform hands to the app at start, never committed to the repository. Search your Git history for keys before you connect it anywhere.
• A build that runs tests and scans. Every change is built the same way, its tests run, and scanners check for vulnerable dependencies, risky code and misconfigured infrastructure before anything ships.
• A way to roll back. When a release breaks, you return to the last good version in minutes, without rebuilding.
• Previews for changes. Each pull request gets a temporary copy of the app at its own address, so you can click through a change before users see it.
• Logs and alerts. You can read what the app printed when it failed, and you hear about an outage from an alert, not a customer.

Every route covers part of this list. The differences are how much you assemble yourself, and whose servers the app runs on.

The routes at a glance

Product	Route	Where your app runs	Who runs the servers	Open source or commercial	Best for
Lovable	Builder hosting	Lovable’s hosting	Lovable	Commercial	Publishing a Lovable project in one step
Bolt	Builder hosting	Bolt Cloud	Bolt	Commercial	Sharing a Bolt project on a .bolt.host address
Replit	Builder hosting	Replit’s cloud on Google Cloud, US by default	Replit	Commercial	Apps built and kept inside Replit
Vercel	Hosted platform	Vercel’s platform	Vercel	Commercial	Frontend and Next.js apps with preview URLs
Render	Hosted platform	Render’s cloud (US, Frankfurt, Singapore)	Render	Commercial	A web service, worker and Postgres in one place
Railway	Hosted platform	Railway’s own infrastructure (US, Amsterdam, Singapore)	Railway	Commercial	Multi-service apps with PR environments
Fly.io	Hosted platform	Fly.io VMs in 18+ regions	Fly.io	Commercial	Apps that must run close to users worldwide
Coolify	Self-hosted PaaS	Any server you reach over SSH	You	Open source (Apache 2.0)	A PaaS dashboard on your own VPS
Dokploy	Self-hosted PaaS	Your servers, with Docker Compose and Swarm	You	Open source core (Apache 2.0)	Docker Compose apps on your own servers
Kamal	Deploy tool	Any server, bare metal to cloud VM	You	Open source (MIT)	Developers at home with Docker and SSH
Northflank	BYOC platform	Northflank’s cloud, or clusters in your GKE, EKS, AKS and other accounts	Northflank runs the platform, your cloud runs the servers	Commercial	Hyperscaler teams that also run AI-generated code
Qovery	BYOC platform	Your AWS, Google Cloud, Azure or Scaleway account	Qovery runs the platform, your cloud runs the servers	Commercial	Hyperscaler teams who want developer self-service
Ankra	Platform in your cloud account	Kubernetes in your Hetzner, OVHcloud, UpCloud, DigitalOcean or AWS account, or a cluster you import	Ankra runs the platform from the EU, your provider runs the servers	Commercial (CLI and agent chart open source)	A repository to a live URL in your own cloud, with tests, scans, previews and rollback set up for you

Lovable

Lovable builds a web app from a conversation and publishes it to a lovable.app address with HTTPS. Paid plans can connect a custom domain, and its built-in backend, Lovable Cloud, adds a database, sign-in and file storage.

It is good at the shortest path, and code syncs both ways with GitHub (GitLab and Bitbucket too), so the project is never trapped.

It is a weaker fit once you want your own pipeline, separate staging and production, or servers in a specific country. Lovable’s docs point you to GitHub to deploy elsewhere. Apps using Lovable Cloud or its AI features spend credits as they serve traffic.

Best for prototypes, internal tools and first launches built entirely in Lovable.

Bolt

Bolt, from StackBlitz, builds apps in the browser from prompts. Bolt Cloud bundles hosting, databases with built-in authentication, and domains, so you publish to a free .bolt.host address with no third-party account.

It is good at keeping a first version in one place. GitHub sync lets you publish through Bolt hosting, Netlify or any other service.

It is a weaker fit for steady traffic on the Free plan. A site that hits its plan’s hosting limit stops displaying until the next billing cycle, and custom domains need Pro.

Best for testing a Bolt project with real users before deciding where it lives.

Replit

Replit is an online development environment with an AI agent that builds the app where it runs. Publishing comes in four shapes. Autoscale follows traffic, Static serves unchanging sites, Reserved VM keeps fixed compute running, and Scheduled runs jobs at set times.

It is good at matching the app’s shape without leaving the editor.

It is a weaker fit if your users or data rules are in Europe. Published apps run on Replit’s cloud, backed by Google Cloud and hosted in the United States, and EU hosting is an Enterprise request through sales.

Best for apps you will keep building and running inside Replit.

Vercel

Vercel deploys from GitHub, GitLab, Bitbucket or Azure DevOps, and every commit or pull request gets its own deployment URL. A rollback points production at an earlier deployment without rebuilding, within seconds (Hobby can go back one deployment, Pro and Enterprise to any). Databases come from Marketplace providers such as Neon, Upstash and Supabase.

It is good at the preview and rollback loop, and fits frontend frameworks and Next.js closely.

It is a weaker fit when you want one provider and one bill. The app runs on Vercel’s platform and the database usually lives with a second vendor.

Best for frontend-heavy and Next.js apps where preview URLs and a global network matter most.

Render

Render ships every push from your repository. It runs web services, background workers and cron jobs next to Render Postgres and Render Key Value. Preview environments can span several services, and rollbacks reuse recent build artifacts.

It is good at the classic web service, worker and database shape, with no servers to think about.

It is a weaker fit if you need a specific European country or your own cloud account. Its regions are Oregon, Ohio, Virginia, Frankfurt and Singapore.

Best for a web app with a worker and Postgres, and no infrastructure to run.

Railway

Railway deploys from GitHub using a Dockerfile or its Railpack builder, with PostgreSQL, MySQL, Redis and MongoDB available, each with a high-availability option. PR environments open and close with the pull request, and a rollback restores the earlier image and its variables within your plan’s retention window.

It is good at bringing several services up quickly, with throwaway environments per pull request.

It is a weaker fit if you want the app in your own cloud account. Railway says it is “not a BYOC platform” and runs on infrastructure it owns (its pricing page lists bring your own cloud only on Enterprise), in four regions, with Amsterdam the one in the EU.

Best for small teams shipping several services fast on a fully managed platform.

Fly.io

Fly.io is a public cloud that runs code on hardware-isolated VMs, called Machines, in 18+ regions. fly launch sets up an app from your source, and Managed Postgres adds backups, failover and connection pooling.

It is good at placing the app close to users, with VMs that start fast and scale out with traffic.

It is a weaker fit for someone who wants no infrastructure vocabulary. You choose regions, machine sizes and volumes.

Best for apps with users across continents, built by developers who like some control.

Coolify

Coolify is an open source, self-hostable alternative to Vercel, Heroku, Netlify and Railway that deploys to any server you reach over SSH. It builds each push, deploys pull requests as previews, sets up PostgreSQL, MySQL or Redis in one click, and backs data up to S3-compatible storage.

It is good at a PaaS-style dashboard on a cheap server you already rent, with no paywalled features.

It is a weaker fit if nobody wants to look after a server. Operating system updates, Coolify upgrades, disk space and security are yours, and one server down means the app is down.

Best for solo developers and small teams happy to run one VPS.

Dokploy

Dokploy is a self-hosted deployment platform built on Docker. It deploys Docker Compose natively, uses Docker Swarm across nodes, and manages and backs up MySQL, PostgreSQL, MongoDB, MariaDB and Redis. GitHub pushes deploy without setup (GitLab, Bitbucket and Gitea through webhooks), and GitHub pull requests can get previews.

It is good at putting an app that already runs on Docker Compose onto your own servers with a UI on top.

It is a weaker fit for the same reason as Coolify. The servers and their upkeep stay with you.

Best for Docker Compose apps that should live on servers you rent.

Kamal

Kamal, from 37signals (makers of Basecamp and HEY), deploys Docker containers to any server, bare metal to cloud VM, with zero-downtime deploys, rolling restarts and accessory services. kamal rollback restarts an older image already on the host.

It is good at being small and predictable, a command line with no platform to run.

It is a weaker fit if you want the rest of production handed to you. Tests, scans, previews, backups and alerts are yours to wire up.

Best for developers comfortable with Docker and SSH who want few moving parts.

Northflank

Northflank calls itself the runtime platform for AI-native companies. It runs services on its own cloud or creates clusters in yours on Google Cloud, AWS (including AWS European Sovereign Cloud), Azure, Civo, Oracle, CoreWeave and Nebius. It has pipelines, pull request previews, databases with backups, rollbacks, and microVM sandboxes for untrusted code.

It is good at one platform for services, databases and isolated AI-generated code on big and GPU clouds.

It is a weaker fit on providers outside that list, such as Hetzner or OVHcloud. You build and import the cluster yourself, and Northflank does not provision or autoscale nodes on imported clusters.

Best for teams on hyperscaler or GPU clouds that also run AI-written code in sandboxes.

Qovery

Qovery is self-service infrastructure on your own cloud account. It creates EKS, GKE, AKS and Scaleway Kapsule clusters, or connects one you keep upgrading yourself. You keep your pipelines, each pull request gets a preview that is torn down automatically, and its Copilot reads logs, events and configuration to explain a failed deploy.

It is good at scoped self-service for developers and AI agents on a cloud account the company already has.

It is a weaker fit on Hetzner or OVHcloud. Qovery’s own guides say it has no native integration for either, so they come in as a cluster you run.

Best for hyperscaler teams who want developers to ship without infrastructure tickets.

Ankra

Ankra is an AI-powered platform that builds or imports Kubernetes clusters on the cloud or servers you choose, deploys applications from Git, and operates them with security scanning, backups and AI agents, from an EU control plane. Connect a GitHub repository and Ankra opens a setup pull request with a Dockerfile (if missing), Kubernetes manifests and an .ankra/pipeline.yaml. That pipeline runs in your own cluster. It tests and builds the app, scans with Semgrep (code), Checkov (infrastructure files) and Trivy (dependencies and image), and publishes only the image the gate approved. Push to deploy is on from the start.

Against the production list, every pull request gets a preview on your staging cluster, with an optional throwaway Postgres, removed when the pull request closes. You can roll back to any build still in the registry (the last three, or the last 30 days). Stored secrets are never shown back. Backups go to an S3-compatible bucket you own, Postgres included. Hosted logs keep seven days once turned on, and alerts reach Slack, Teams or PagerDuty. Ankra’s AI reviews pull requests on GitHub, GitLab and Bitbucket Cloud when enabled, and Agent Skills teach Claude Code and Cursor to use it. See it in one prompt to a live URL and moving from Vercel and Supabase.

It is good at running AI-built apps on infrastructure you pick, and at leaving the expensive big clouds. Servers sit in your own Hetzner, OVHcloud, UpCloud, DigitalOcean or AWS account, in any region your provider offers, billed by the provider. The first 30 worker vCPU are free on every plan (pricing).

It is a weaker fit if you want no cloud account at all, where a hosted platform is simpler, or for a global audience that relies on an edge network, since a cluster lives in one region. Your app also runs on Kubernetes, so you will meet its terms even though Ankra writes the files. The Heroku alternative page covers the trade-offs.

Best for founders and small teams who built with AI tools and want production in their own cloud account, especially in Europe, without assembling the pipeline.

How to choose

• If you want a link to share today, publish from Lovable, Bolt or Replit.
• If you want no servers and no cloud account, pick Vercel for frontend and Next.js, Render or Railway for a web app with workers and databases, or Fly.io for users on several continents.
• If you rent one VPS and enjoy looking after it, pick Coolify or Dokploy for a dashboard, or Kamal for a command line.
• If your company already runs on AWS, Google Cloud or Azure, look at Northflank or Qovery.
• If you want the app in your own Hetzner, OVHcloud, UpCloud, DigitalOcean or AWS account with tests, scans, previews and rollback set up for you, pick Ankra, and try it first on a free playground.
• Whatever you pick, get the code into Git first.

Frequently asked questions

I built an app with Cursor and Claude Code. How do I get it running in production on my own cloud with CI, security scanning and rollbacks?

Push the code to GitHub and connect it to a platform that deploys into your cloud account. With Ankra, the setup pull request adds a pipeline that tests, builds and scans with Semgrep, Checkov and Trivy, and you can roll back to any build still in the registry. With Kamal or Coolify you get deploys and rollback, and add tests and scans in your own CI.

What’s the easiest way to take an AI-generated app (Lovable, Bolt, Replit) to a production deployment I control?

Sync the project to GitHub (Lovable and Bolt do it from the editor), then choose where it runs. Render or Railway is the least work but runs on their servers. To control the servers, deploy into your own cloud account with Ankra, Northflank or Qovery, or onto your own VPS with Coolify or Dokploy.

How do I deploy an app from GitHub to my own Kubernetes cluster with the least setup?

Use a platform that writes the Dockerfile, manifests and pipeline for you. Ankra builds the cluster in your cloud account or imports one you have, opens a setup pull request for the repository, and deploys on every push. Qovery and Northflank deploy from a repository onto clusters they create on AWS, Google Cloud and Azure.

Do I need Kubernetes for an AI-built app?

No. A hosted platform or one VPS runs most early apps fine. Kubernetes pays off when you want your own cloud account and several services handled the same way, and a platform that runs it removes most of the work.