claude code

A corporate VPN for teams that work with Claude Code

Claude Code runs long sessions in the terminal, and it works best on a connection that does not change under it. Guard Tun gives the whole team that connection: one route, and an egress address that belongs to your company.

the problem

What Claude Code needs from the network

An agent session is not a single request. Claude Code reads files, runs tests, edits code and talks to Anthropic's services for as long as the task takes, which can be many minutes and sometimes much longer. During all that time it needs the same network path to stay in place.

A connection that switches exit addresses halfway, or drops for a moment, costs work: a half-finished task, a retry, a session somebody has to restart. On a consumer VPN the exit address is also shared with strangers, and you do not choose it. For a team, the useful thing is a route that every developer has, that does not move, and that your administrators control.

routing

One route, one address

The client on each machine recognises the Anthropic domains and sends that traffic through an encrypted tunnel to a node on your own server. The node is where the traffic leaves from, so the address Anthropic's services see is your server's address, the same for every developer on the team.

Only the listed domains take this route. Package registries, video calls, the browser and everything else go direct, exactly as before, so nobody pays for the VPN in everyday speed. The nodes and the console run on your servers, not on ours.

leak protection

If the tunnel drops

Reliable leak protection is part of the client on Windows, macOS and Linux. If the tunnel goes down, traffic to the listed domains stops instead of going out directly from the developer's own address. The session pauses rather than leaving the company through a path you did not choose.

Meanwhile the client works to bring the tunnel back. If another VPN on the machine is in the way, the client names it and the operator gets a hint about what to fix.

linux and ci

In the terminal and on servers

desktop

Developer machines

Native clients for Windows and macOS, installed from the profile link the administrator sends.

linux

Build servers and workstations

A Linux client brings the same route to remote development machines and headless runs of Claude Code.

automation

Scripts and runners

Devices and runners are provisioned with a key. Command-line tools return exit codes, so a pipeline can check the result of each step.

console

For the administrator

The console shows who is connected, on which client version, and the live status of every node and every device. When a developer says Claude Code feels slow or has stopped, you can look at the device instead of guessing.

Clients update themselves from a signed manifest, and a version that fails is rolled back, so the route does not depend on anyone remembering to upgrade.

faq

Questions

Does Claude Code need any setting changed?

No. Routing is by domain on the device, so Claude Code keeps working as it does today. The client sends the Anthropic domains through your node and nothing in the tool has to be configured.

Does it cover other tools at the same time?

Yes. The same client routes the domains of Cursor, OpenAI and JetBrains too, so a developer who uses several AI tools has one route for all of them. Everything else goes direct.

Can we use it for Claude Code in CI?

Yes. A build runner or a headless machine is provisioned with a key and uses the Linux client, so it takes the same route as the developers. Command-line tools return exit codes, so a script can check the result.

Is this an Anthropic product?

No. It is an independent product that works with Claude Code. The footer of every page carries the trademark notice.

request

Request a demo or a pilot

Tell us about your team, and we will come back with a time for a call and a plan for a pilot.