Your code stays on your machine.
Command Center is a local-first Windows app. Sessions run on your PC and talk to Anthropic with your own key. This page lists what leaves the machine, what never does, and how to report a problem to us.
Last updated 3 November 2026. Questions: use the contact form and choose Security.
Local first, drawn as it runs.
Everything inside the outlined box is your PC. Each line out of it is a connection you can name, and three of them exist only when you turn the feature on.
Paired device. Sees questions and status, sends answers.
⇄Tailscale, private network onlyPlanned for Pro and above. Settings and session history.
⇄planned sync, no API keysNever leaves this box: your API keys, your code (except what a session sends to Anthropic), your transcripts.
Model requests made by your sessions, billed to your key.
⇄HTTPS, your own keyTeam and above. Only the sessions you run for a company you share, after redaction.
⇄HTTPS, redacted transcriptsWhen accounts open, the app will contact it with your email, plan and member count, for sign-in and licence checks.
⇄HTTPS, no code, no transcriptsYour API key: encrypted, never logged.
You add your Anthropic API key in Settings, then Accounts. From that moment it lives in one place on your PC and appears nowhere else.
- Encrypted on device. Keys are stored encrypted with your Windows account, so another user on the same PC cannot read them.
- Read only when a session starts. The key is passed to that session's process and to nothing else.
- Never logged or rendered. Keys are stripped from logs and from child processes that do not need them. The app shows the last four characters only.
- Never copied between PCs. You add keys on each machine.
- Never sent to us. The account service we will open with accounts has no field for an API key.
Settings, Accounts
What leaves your machine, and what never does.
The full list. If a connection is not in the left table, Command Center does not make it.
leavesWhat leaves, and where it goes
Each row is a connection you can switch off, except the first and the last.
| Prompts, file contents and tool results a session sends | To the Anthropic API, with your key, under your agreement with Anthropic. |
|---|---|
| Questions, status and your answers | To your paired phone, over Tailscale. Only if you turn on Remote. |
| Settings and session history Planned | Planned: to your second PC, on Pro and above, if you turn on sync. Not in the early-access build. |
| Sessions you run for a company you share | To your Team Hub, after redaction on your PC. Only on Team and above. |
| Email, plan, member count, app version | When accounts open, to our account service, for sign-in, licence checks and update notices. |
staysWhat never leaves
These stay on your PC whatever you turn on.
- Your Anthropic API keys.
- Your code and repositories, apart from what a session sends to Anthropic to do its work.
- Transcripts and reports, unless a session is uploaded to your Team Hub.
- The contents of review pages and your notes on them.
- Usage analytics about what you type or which files you open. We do not collect any.
Remote, redaction and checkpoints.
The three features that reach beyond one session each have a boundary built in.
-
Phone remote over a private network
The remote listens on your PC's Tailscale address and on 127.0.0.1, and on nothing else. It is never bound to a public interface, and no port is opened to the internet.
- A phone must be on your Tailscale network and paired with a one-time code.
- A request from the PC itself gets no special trust.
- Remove a paired device from Settings and it loses access at once.
-
Team Hub redaction
Sessions you run for a company you share are uploaded to the Hub automatically. Before upload, keys, tokens and passwords in the transcript are replaced on your PC.
- Token-shaped strings such as API keys, bearer tokens and private keys.
- Credentials inside URLs, such as git remotes.
- The value after a label such as token, secret, password or api key.
-
Permissions and checkpoints
In a coordinated job, a command that sends to an audience, starts a paid generation or deploys stops at a checkpoint and waits for you. Other sessions use Claude Code's own permission prompts.
- A held command shows the session as waiting until you answer.
- Approve lets the command run; Change sends the job a note; Cancel stops it.
Responsible disclosure.
If you find a security problem in Command Center, the account service or this website, tell us through the contact form with the topic set to Security. Please include the steps to reproduce it, the version you tested and how we can reach you.
- We will not take legal action against research done in good faith that follows this policy.
- Do not access other people's data, degrade the service or test social engineering against our staff.
- Give us 90 days to fix the problem before you publish. We will agree a date with you and credit you if you want.
Subprocessors.
The companies we use to run this website, the account service and email. None of them receives your code, transcripts or API keys.
No payments are taken yet. Paid plans will be billed through a payment provider that we name here before billing starts. We notify customers 30 days before adding a subprocessor.
| Subprocessor | Purpose | Data | Location |
|---|---|---|---|
| Vercel Inc. | Hosts this website and stores the messages sent through its forms | Name, email, form messages | United States |
| Resend | Delivers the messages sent through this website's forms to our team | Name, email, form messages | United States |
- Anthropic is your provider
Model requests go from your PC to Anthropic under your own agreement with them. Anthropic is not our subprocessor for your data.
- Compliance
SOC 2 is on our roadmap. We do not hold a certification today and will say so here until we do.
- Legal documents
Read the privacy policy, the data processing agreement and the terms.
- Security review
Enterprise customers can request our security questionnaire answers through the contact form (topic Sales).
Questions before you install?
Send a message to our security contact, or read the data processing agreement first. A person answers every message.