Deploy and host apps built with Claude Code
From a Claude Code project folder to a live URL, including the one network check that matters in a remote environment.
Claude Code hosting ↗WORKS WITH YOUR AGENT
Your agent
needs two things.
Most have both.
You are not the one who deploys a Saasie — the agent you are already talking to does it for you. So whether this works has nothing to do with which model you like, and everything to do with what that agent is allowed to do on your behalf.
YOUR AGENT
SANDBOX
Let it reach the internet
Off in a lot of hosted sandboxes, by default
- Allowed
Which hosts it may reach
Four of them, if you get to pick
Two switches. Your agent will call them something else, and a good few have them on already.
A computer of its own ✦ A way out to the internet ✦ That is the whole list
CHOOSE THE TOOL IN FRONT OF YOU
Your agent.
Its exact route.
The hosting is the same. The useful difference is where each agent runs, how it gets a terminal, and whether its network is already open. Pick yours for the short version written around that setup.
Deploy apps built with OpenAI Codex
The route for Codex on your computer and Codex in a cloud environment, where internet access is a deliberate permission.
Codex app hosting ↗Host apps built with Cursor
Let Cursor Agent use the terminal it already works in, then have it package, push and verify the app without leaving the editor.
Cursor app hosting ↗Vibe coding hosting
A tool-neutral guide for taking code from any capable AI coding agent and turning it into a running app with its own address.
Compare the route ↗WHAT IT ACTUALLY NEEDS
Two things,
and that is all.
Somewhere to run a program
Saasie is a small tool your agent runs, the same way it runs anything else — so it needs a machine of some kind to run it on. Your laptop counts. So does a sandbox rented by the minute in somebody's data centre. An agent that can only write text back into a chat window does not have one, however good the code in that text is.
A way out to the internet
Deploying means reaching us: sending the code, watching the build, opening the finished app to see that it works. Hosted sandboxes often start with the network switched off, which is a sensible default and the single most common reason a capable agent tells you it cannot do this. It is a setting, not a limitation.
FIND YOURS
Three kinds
of agent.
Sorted by what they can do rather than by who makes them, because the same company will happily sell you one of each.
NOTHING TO DO
The ones on your own computer
They already have a machine and a network — yours. Ask for a Saasie and they get on with it.
- Claude Code
In the terminal, the desktop app, or your editor.
- Codex CLI
The one you run yourself, not the one in the browser.
- Gemini CLI
Same again: on your machine, it can do this.
- Cursor · Windsurf · Zed
Editor agents, once they are allowed to run commands.
- Anything with a terminal
Including an agent your company runs behind its own gateway.
ONE SETTING
The ones in a sandbox somewhere else
A real machine, rented for the length of the conversation, and usually walled off from the internet until you say otherwise.
- Claude, on the web
In settings, under capabilities: allow network egress, and set the domain allowlist to all domains.
- Claude Code on the web
Allow the four hosts below in the environment's network settings.
- Codex, in ChatGPT
Its cloud tasks start with no network. Turn it on for that environment.
- Copilot coding agent
It works behind a firewall somebody with admin on the repository can open.
- Kimi
Its sandbox is already open. Nothing to change.
NOT THIS WAY
The ones that are only a chat window
Nothing to run a program on, so nothing to deploy from. This is not a permission you can grant — there is no machine to grant it to.
- Gemini in the browser
Writes the app happily. Cannot put it anywhere.
- ChatGPT, plain chat
Its sandbox has no way out to reach us.
- Le Chat · DeepSeek · the rest
Same shape, same answer.
If the same product also offers an agent mode with a terminal in it, that mode belongs in the middle group and this one does not apply. And a conversation like this is not wasted: keep the code it wrote, hand the folder to an agent from either group above, and ask that one to put it live.
Checked August 2026. These are other people's products and they change their sandbox rules often — if this page is wrong about yours, tell us at hello@saasie.io. We would rather know than be trusted.
IF YOU GET TO PICK THE HOSTS
Four addresses,
nothing else.
Some agents let you name the hosts instead of opening everything. This is the whole list — worth handing to whoever keeps the allowlist at work.
- saasie.io:443 the tool itself, and the instructions your agent reads
- api.saasie.io:443 the platform — every command your agent runs
- git.saasie.io:2222 where your agent sends the code
- *.saasie.app:443 your finished app, so your agent can check it works
The install page has the same list next to the tool it belongs to. If saasie.io itself is blocked there is no second place to get this from, and a decent agent will say so plainly rather than improvise around it.
WORTH BEING PLAIN ABOUT
Where the
edges are.
- Opening the network is a real decision
- A sandbox that cannot reach the internet cannot leak anything through it either, and that is the point of the default. Letting your agent out is a permission you are granting for the whole conversation, not only for us — which is why the four hosts above exist, for anyone who would rather name them than open the door.
- These settings are not ours to move
- We do not build any of these agents and cannot change what their sandboxes allow. What we can do is keep the list of hosts small enough to be allowed, and keep this page honest about which agents currently work.
- Choosing Saasie is not choosing an agent
- Nothing here is tied to one model or one company, and you can change your mind next month. A Saasie already running does not care which agent made it, or which one you bring along to change it next.
ONE SENTENCE IS THE WHOLE JOB