GPU jobs are being built
Apps and AI,
in one project.
Deploy apps from your repo today. GPU jobs, model endpoints, and an AI gateway are coming to the same project.
Stack
Bring any stack.
Your stack is detected and built automatically. Anything else ships with its Dockerfile.
Detected Python · no config needed
- Python
- Node.js
- Go
- Rust
- PHP
- Rails
- Django
- Bun
- Nuxt
- Spring Boot
- SvelteKit
- Flask
- Astro
- Phoenix
- Deno
- React Router
- Express
- Hono
- Dockerfile
- Next.js
- FastAPI
- Vite
- Python
- Node.js
- Go
- Rust
- PHP
- Rails
- Django
- Bun
- Nuxt
- Spring Boot
- SvelteKit
- Flask
- Astro
- Phoenix
- Deno
- React Router
- Express
- Hono
- Dockerfile
- Next.js
- FastAPI
- Vite
Push code. Get a live app.
Connect a repository and every push becomes a live release. Each pull request gets a preview, and a release that never gets ready rolls back on its own.
Live from a repo. Connect GitHub or paste a public repo link, then ship with a git push, a message to your AI agent, or one CLI command.
- #47Faster product searchmain · 9e1b0d4 · via git push Live
- #46New pricing pagemain · 51c0a7e · via CLIReplaced
- #45Update footer linksmain · 0b7d3c2 · via agentReplaced
Any language. Node.js, Python, Go, Ruby, PHP, Java, Rust, and more are detected and built for you. Anything else ships with its Dockerfile.
Detected Python · no config needed
Safe releases. A release that never gets ready rolls back on its own, and its env vars roll back with it. Canary releases send a slice of traffic first and roll back when errors rise.
- Building
- Canary gets 10% of traffic
- All traffic to #48
- Old version stops
Preview every pull request. Each pull request gets its own live URL, posted on the pull request and redeployed with every push.
Adds a wishlist button to the product page.
4f72aa1…Visit preview URL · View deployment logs
Live is only the start.Then it has to keep running.
Everything your app needs to run.
Your apps reach each other over a private network, never over the internet. When something breaks, the error comes with the fix.
Workers, jobs, and Postgres. Background workers, cron jobs, and managed PostgreSQL, in the same project and on the same network.
Private networking. Cubes in a project reach each other over a private address that never touches the internet.
Only web is on the internet · api, worker, and postgres stay inside the project
Scales to zero. An app with no traffic sleeps and frees its memory. The next request wakes it.
Errors that tell you the fix. A failed deploy shows its code and the fix. AI explains runtime errors.Soon
Nothing is listening on $PORT. Listen on the port Cubecen gives you.
- app.listen(3000)
+ app.listen(process.env.PORT)
Close to your users. Run each app in the region nearest its users, so every request travels less.
to Virginia, the nearest region
Your AI, next to your apps.
Train on a GPU, serve what you trained, and route every AI call through one place. All of it in your app's project, with the same env, secrets, and logs.
GPU jobs.Soon Add a job like python train.py to your repo, then press Run or set a schedule. It runs on a GPU, streams its logs, and bills only the minutes it ran.
jobs: train: command: python train.py
Pulling acme/support-bot@9e1b0d4...
Same env, secrets, and database as your app · logs next to your app's
Models become endpoints.Direction Serve the model a job trained and get an address. It wakes on the first request and sleeps when traffic stops.
POST https://support-model-acme-91c2d0.cubecen.app/v1/chat/completions
No traffic · nothing billed
Asleep costs nothing · illustrative
One gateway for every AI call.Direction OpenAI, Anthropic, and your own models behind one address. Keys stay in Cubecen, every call shows its cost, and a budget stops spend at the limit.
- support-model380 ms$0.0002#48
- claude-sonnet1.1 s$0.0031#48
- gpt-4o-mini412 ms$0.0004#48
Keys stay in Cubecen · calls stop at the limit · illustrative
Everything you need to know.
Is Cubecen free?
Yes. The Free plan runs up to 3 small cubes with deploys from GitHub or a public repo link, releases and rollback, automatic HTTPS, metrics, and logs, with no credit card. Launch and Scale have no plan fee: you pay for the compute your cubes use, by the vCPU-hour, plus bandwidth and build minutes beyond what's included. During early access, Launch and Scale are free for approved accounts; request the plan from your dashboard. Enterprise covers custom limits and terms. Every plan is on the pricing page.
What can I deploy?
Web apps and APIs in almost any language. Your stack is detected and built automatically, and anything else deploys with its Dockerfile. Each app is a cube with its own HTTPS address.
Which parts are live today?
Apps from GitHub or a public repo link, env vars and write-only secrets, private networking, custom domains with HTTPS, logs, metrics, rolling releases, and automatic and one-click rollback. GPU jobs are being built. Model endpoints and the AI gateway are on our roadmap.
How will GPU jobs work?
You add a job to your repo, for example python train.py, and press Run or give it a schedule. It runs on a GPU with your app's env and secrets and bills the minutes it ran. GPU jobs are being built.
Do you look at or train on my data?
No. Cubecen doesn't monitor or track the data that goes through your apps, jobs, or models, and never uses it to train anything of its own.
What happens if a deploy fails?
Your running release keeps serving. New replicas have to accept connections on $PORT before the old ones stop. If they never do, the release rolls back automatically and you get a failure code with a suggested fix.
Ready to ship?
Push a repo and your app is live. Free to start, no credit card. GPU jobs come next.