Security
Last updated November 18, 2026
Cloud Coding runs an AI agent on a copy of your code. That asks for trust. This page says in plain words how we protect your code and your accounts, what the agent can and cannot do, and what you should still watch yourself. For the data we collect and who gets it, read the privacy policy.
At a glance
- Your own VM. Your code runs on a cloud computer that only you use.
- Closed to the internet. Nothing on the internet can connect to your VM.
- Encrypted at rest. The disk of your VM and your stored project files are encrypted.
- An agent with narrow access. It has no cloud keys. It reaches only your VM, for one chat at a time.
- Secrets stay encrypted. We encrypt your GitHub token and your own API keys. We never see your card number.
- No passwords. You sign in with GitHub.
- You can leave. Delete your account, and your VM, files and chats go with it.
Your own VM
When you join, we make a VM for you. It is a separate virtual machine, not a container on a machine that other users share. Your projects are copied to your VM and to no other.
- Nobody else's code runs on your VM. Yours does not run on theirs.
- Nothing on the internet can start a connection to your VM. Only our API server can.
- VMs cannot reach each other.
- Your VM can reach the internet. The agent needs it for things like
npm installandgit pull. - Your VM has no cloud account keys. A program on it cannot use our cloud account.
- The disk of every VM that we make is encrypted. It is deleted with the VM.
- An idle VM stops by itself. It starts again when you send a message.
Who can reach your VM
| Who | Access | How |
|---|---|---|
| You | Yes | Through the apps, or over SSH with your own key. Your key reaches only your own VM. |
| The AI agent | Yes, during a reply | It sends commands through our API. It cannot connect to the VM by itself. |
| Our API server | Yes | It runs your requests: the agent's commands, copying your projects, and starting and stopping the VM. |
| Other users | No | Not their agents and not their keys. |
| The internet | No | We block inbound connections. |
We run the API server, so we can reach any VM through it. We use that access to run your requests and to fix problems that you report. We do not browse your code.
What the agent can and cannot do
The agent can:
- Run commands on your VM, and read and edit your project files there.
- Use your GitHub access to commit, push, or make a repository when you ask. In a project that you made in the cloud, it commits and pushes by itself.
- Search the web and read web pages.
- Use your MCP servers, if you turned on that sync.
The agent cannot:
- Reach another user's VM or data.
- Use our cloud account. It has no cloud keys and no SSH key.
- Use our API for more than its own chat on its own VM. Its token stops working when the reply ends.
- Change files on your Mac. It works on a copy. Changes come back only when you pull them.
One risk to know. Like any coding agent, it can be misled by text in a file, an issue or a web page (prompt injection). Your VM is the boundary. A misled agent can use what is on that VM: your code and your GitHub access. It cannot reach other users' data or our cloud account. So:
- Keep out of your synced projects the secrets that you do not want in the cloud.
.envfiles sync only when you turn that on, or when git tracks them. - Look at what the agent pushes. Use branches and pull requests for work that matters.
Your accounts and secrets
- You sign in with GitHub. We store no passwords. The apps use OAuth with PKCE.
- We encrypt your GitHub token and your own Anthropic and OpenAI keys (AES-256-GCM) before we store them. The dashboard shows only the last 4 characters of a key.
- We hide your GitHub token in the output of the agent's commands, so it does not reach your chat or its traces.
- Stripe takes your card. Cloud Coding never sees the number.
- You can remove Cloud Coding from your GitHub settings at any time.
The website and the connection
- Everything goes over HTTPS, with HSTS.
- The pages of cloudcoding.ai run no JavaScript. The browser enforces this with a Content-Security-Policy, so a script that someone injects cannot run.
- Our session cookie is
HttpOnly,SecureandSameSite. A form that changes something checks where the request came from.
How we run our servers
- The API server runs as an unprivileged user, in a restricted sandbox.
- Its cloud permissions are narrow. It can make, start, stop and delete only Cloud Coding VMs, from one approved image. It cannot do other things in our cloud account.
Your controls
- You choose which projects sync from your Mac.
.envfiles stay off the sync unless you turn them on.- MCP server sync is off until you turn it on.
- Sending your Claude Code chats to Cloud Coding is on by default. You can turn it off in the Mac app's settings.
- You can use your own Anthropic or OpenAI key, so the model calls go straight to that provider.
Who else sees your code
To write replies, the AI providers get your messages and the parts of your projects that the agent reads. We do not sell your data. The privacy policy names each service.
Deleting your data
Open Settings in the iPhone app and tap Delete account. We delete your VM and its disk, your chats, your projects and their stored files, your SSH keys, your API keys and your saved card. We also remove Cloud Coding's access to your GitHub account. We keep our payment records. Your GitHub repositories stay yours. We delete your LangSmith traces when you ask. The privacy policy has the full list.
What we do not have yet
We are a small team. We have no third-party security audit or certification, such as SOC 2. We would rather tell you how Cloud Coding works than show you a badge.
Report a problem
Did you find a security problem? Write to sean@lmsystems.ai with "Security" in the subject. Tell us what you found and how to repeat it. Please test only your own account and VM, do not read other users' data, and give us time to fix the problem before you share it. We read every report and we will reply.