A real server for your AI agent, in one HTTP call.
Coding agents are good at writing software and bad at the part where it has to run somewhere. They need a machine they can SSH into, install things on, keep state on, and point a URL at — without a human clicking through a dashboard. That's what this is: an isolated Linux server an agent can create, use and ship from on its own.
One call, a whole server
curl -sX POST https://thevibehosting.com/createNo signup, no key to configure, no MCP to install. The response gives the agent everything it needs:
sshandssh_private_key— root on an Ubuntu box,subdomain— a live HTTPS URL that already serves a page,api_key— the account credential for everything else (rename, custom domain, sleep, delete, billing),ONBOARDING— a short guide the agent can read straight from the response.
The API is self-describing: errors say what to do next, and SKILL.md / llms.txt explain the whole workflow in a form agents read well.
A sandbox you don't have to trust the code in
- gVisor isolation — every server runs in its own sandboxed kernel, separated from the host and from other tenants.
- Network isolation — a server can reach the internet, not other tenants or our internal network.
- Hard limits on CPU and RAM per server, and automatic suspension of abuse such as scanning or mining.
- Outbound email (SMTP) is blocked, so a compromised project can't send spam.
What agents use it for
- A persistent workspace — install toolchains once, keep data, come back to it next session.
- Preview environments — one server per branch or experiment, deleted when done.
Sleeps when idle, or stays on
By default a server scales to zero after about 10 minutes without incoming traffic and wakes on the next request, so idle experiments cost next to nothing. With a verified phone, any server can be marked always-on instead.
Pricing
| Anonymous (no signup) | Verified (phone) | |
|---|---|---|
| Free credit, every month | $0.90 | $2.25 — a 512 MB server runs 24/7 free |
| RAM per server | up to 256 MB | up to 2 GB (4 GB once you hold paid credit) |
| CPU | up to 2 cores, always free | |
| Always-on (never sleeps) | no — sleeps when idle, wakes on the next request | yes, optional per server |
| Your own domain | no | yes, up to 3 per server, HTTPS automatic |
| Servers | 2 per network | unlimited |
| Credit card | never required to start | |
After the free credit: RAM $0.005 per GB-hour while the server is awake, outbound traffic $0.10/GB, disk $0.25 per GB-month. No plans, no reserved sizes — you pay for what actually runs.
FAQ
Does my agent need an MCP server to use this?
No. It's a plain HTTP API plus SSH. Any agent that can run curl and ssh can use it; SKILL.md explains the workflow.
Is it safe to let an agent run arbitrary code here?
Each server is a gVisor-sandboxed container with hard limits, isolated from the host and other tenants, and abuse is suspended automatically. Treat the server itself as untrusted from your side too: keep real secrets out of it unless the project needs them.
Which agents does it work with?
Anything that can make an HTTP request and use SSH — Claude Code, Cursor, Codex and others.
Try it: paste this to Claude Code, Cursor or Codex.