You built it with AI. Now ship it like a pro.
Lovable, Cursor, Claude Code, Bolt and Replit get you to a working preview. ComputeSphere gets it live and keeps it up, with your keys out of the code, a health check on every release and a way back when something breaks.
Card required. No charge until day 15.
Ship this to production on ComputeSphere.
Reading AGENTS.md. Checking before I deploy:
Passed: Listens on 0.0.0.0 and $PORT
Passed: Health check at /healthz answers 200
Problem: STRIPE_SECRET_KEY is hard-coded in src/lib/pay.ts
Move it to a secret before we go live? I'll show you the command first.
Yes, then deploy.
It works in the preview. Then it breaks live.
AI tools build for the preview window. Production has real users, real keys and no undo, and the same five gaps show up again and again.
Your keys are in the code
Anyone who can see the repo, the browser bundle or a screenshot now has your payment or AI key.
src/lib/pay.ts Preview: const stripe = new Stripe("sk_live_demo-not-a-real-key-0000")Production: const stripe = new Stripe(process.env.STRIPE_SECRET_KEY)Your data is open to anyone
The database rules were loosened so the preview would load. Live, any visitor can read and change every row.
database.rules Preview: allow read, write: if true;Production: allow read, write: if request.auth.uid == resource.data.ownerId;The app listens on localhost
It answers on your laptop and nowhere else, so the live URL never comes up.
server.js Preview: app.listen(3000, "localhost")Production: app.listen(process.env.PORT || 3000, "0.0.0.0")Nothing checks it's healthy
A release that crashes on start still gets traffic, and you find out from your users.
server.js Preview: // no health routeProduction: app.get("/healthz", (req, res) => res.sendStatus(200))There's no way back
When a change breaks sign-up, the only fix on offer is another rushed change.
terminal Preview: git push --force # and hopeProduction: csph deployments rollback <deployment-id>
The red line is what the preview needed. The green line is what production needs. Learn each fix
Where your app goes live and stays up.
ComputeSphere runs your app and handles the parts your AI tool left out.
Card required. No charge until day 15.
Deploy from GitHub or your agent
Connect a repo and ComputeSphere builds it from your Dockerfile or from source. Or let your coding agent deploy through our MCP server or the csph CLI.
Secrets stay out of your code
API keys and database URLs are stored as secrets on your environment and reach your app as variables. Nothing sensitive goes in the repo.
Health checks guard each release
A new version takes traffic only once it passes its health check. A release that won't start stays away from your users.
One-click rollback
Go back to the last good version from the console, or with one csph command, then fix the bug calmly.
Live logs
Build, deploy and runtime logs for every service, in the console or your terminal. Your agent can read them too.
Custom domains with SSL
Point your domain at your app. Certificates are issued and renewed for you.
Bring your database. Keep it with the hosted provider you already use and pass its URL to your app as a secret. ComputeSphere doesn't host databases or run GPUs.
Ship with your coding agent.
Connect your agent to ComputeSphere in one line. Then give it the rules a careful engineer would follow.
Claude Code
Run in your terminalclaude mcp add --transport http computesphere https://mcp.computesphere.com/mcpCursor
Add to ~/.cursor/mcp.json{
"mcpServers": {
"computesphere": {
"url": "https://mcp.computesphere.com/mcp"
}
}
}The first time your agent connects, your browser opens so you can sign in. Setup for Claude, ChatGPT, Windsurf, VS Code and others is in the MCP server docs.
The agent kit
An AGENTS.md for the root of your repo. Your agent reads it at the start of every session and deploys the way a careful engineer would.
Scoped tokens. A token limited to one project with a short expiry, never your whole account.
Ask before changing live. Deploys, restarts, rollbacks and variable changes each wait for your yes.
Read the logs first. It quotes the log lines its diagnosis rests on, instead of guessing.
Roll back first. If a release breaks, it restores the last good version, then works on the fix.
curl -fsSL -o AGENTS.md https://raw.githubusercontent.com/computesphere-samples/learn/main/agent-kit/AGENTS.mdLearn what your AI skipped.
Free, plain-language lessons on ComputeSphere Learn. Read them before you launch, or after something breaks.
- 8 lessons, about 80 minutes
Ship your AI-built app
Why it works locally and not online, how secrets leak, reviewing an agent's pull request, rollback, and a lab where you ship a starter app.
Open on Learn - Free checklist
Production checklist
13 things to confirm before real users arrive, each linked to the lesson that teaches it.
Open on Learn - Guides
Deploy guides
Step-by-step guides for taking an app from Lovable, Cursor or Claude Code to a live URL.
Open on Learn - 31 lessons, 3 labs
Vibe coding to production
The full path: briefs, reading every diff, tests that prove something, security of AI-written code, and operating with an agent.
Open on Learn
Fair questions.
Something else? Get in touch
Do I need to know how to code?
You don't need to be an engineer, but you will learn a little of what your AI skipped: where secrets go, what a health check is and how to roll back. The Learn track covers it in plain language in about 80 minutes, and your coding agent can do the typing.
Is there a free trial?
Yes, 14 days. A card is required to start, and you're charged from day 15 unless you cancel before then.
Can I bring my database?
Yes. Keep it with the hosted provider your app already uses and pass its connection URL to your app as a secret. ComputeSphere doesn't host databases.
Which AI tools work?
Any tool that gives you a Git repository or a container image, including Lovable, Cursor, Claude Code, Bolt and Replit. ComputeSphere builds from a Dockerfile or from Node.js, Bun, Python, Ruby, Go, Rust or Elixir source. There are no GPUs, so it's for running your app, not training models.
Is my code safe with an agent?
You stay in charge. For unattended work, give the agent a token limited to one project with a short expiry. The MCP server acts as you and asks you to confirm deletions and other large changes, and the agent kit tells your agent to ask before it changes anything live.
Take it from preview to production.
Deploy from GitHub or your coding agent, with secrets, health checks, rollback and logs from the first release.
Card required. No charge until day 15.