The corporate VPN that keeps AI coding tools working
Run it on your own servers. Claude Code, Cursor and Codex reach their services through an address your company owns, and the rest of the traffic goes direct. If the tunnel drops, traffic to the listed AI domains stops instead of going out directly.
- self-hosted
- your egress
- routing by domain
- signed updates
One stable route for the AI tools your developers use
AI coding tools work best on a connection that does not change under them. Guard Tun gives the whole team that connection without putting a third-party VPN provider between your code and the AI service.
An egress address you own
The AI services see one address, and it belongs to your company: the address of your own server, not a shared exit of a consumer VPN that thousands of strangers use.
Only AI traffic in the tunnel
Routing is by domain. Anthropic, Cursor, OpenAI and JetBrains go through your node; video calls, package mirrors and the rest of the web go direct.
Reliable leak protection
If the tunnel drops, the listed domains fail closed: traffic to them stops instead of going out directly, on Windows, macOS and Linux alike.
Your device, your node, the AI service
The client on each machine sends AI-tool traffic through an encrypted tunnel to a node on your server. The node is the only place that traffic leaves from.
-
Put a node on your server
Add the server in the console. It provisions the node over SSH and keeps it maintained from then on.
-
Give each developer a profile
Every device gets its own key and a client for Windows, macOS or Linux, installed from one link.
-
Work as usual
The client routes the AI domains through your node and updates itself from a signed manifest. Nobody has to remember to turn anything on.
Built for the people who run it
Developers get a connection they stop thinking about. The administrator gets a console that shows what is happening and does the routine work.
On your own servers
Nodes and the console run on your infrastructure. You see nodes, devices, usage and the audit log.
auto-updateClients that update themselves
Signed updates that roll back when a new version fails, and the live status of every device.
listsDomain lists that stay current
Your own domains, public sources refreshed every hour, and AI-suggested additions an administrator approves.
scalingAdd nodes as you grow
New nodes are added from the console, and the capacity view shows where the load sits.
securityWhat your security review asks for
Reliable leak protection, encryption, an audit log, signed updates and SIEM export. SSO and MFA for administrators are coming soon.
maintenanceNodes with scheduled maintenance
Repeatable provisioning, drift detection, retries and a maintenance window with a reboot.
automationCI, agents and headless jobs
Build runners and AI agents get the same route through a REST API and provisioning by key.
diagnosticsDiagnostics on every device
The client checks the machine, names another VPN that gets in the way, repairs the tunnel and tells the operator.
platformsWindows, macOS and Linux
Native clients for the desktops your developers use, all managed from one console.
Every node and every device, live
One console for the whole installation. It answers the questions an administrator gets asked, without a terminal.
- Nodes with their reachability and load
- Devices with their status, client version and last connection
- Traffic per user and per node, today, over 7 and over 30 days
- Every change by an administrator in the audit log
Made for the tools your developers already use
VPN for Claude Code
Long agent sessions in the terminal run on one route and one egress address.
cursorVPN for Cursor
The editor, its chat and its agents go through your node; the rest of the developer's traffic does not.
codexVPN for Codex
The Codex CLI and the IDE extension use the same route as the rest of the team.
Why not a consumer VPN, a mesh network or a VPN of your own?
Each solves a different problem. Compared for AI coding tools:
| Guard Tun | Consumer VPN | Mesh VPN | Built by hand | |
|---|---|---|---|---|
| Runs on your servers | yes | no | partly | yes |
| An egress address only your company uses | yes | no | partly | yes |
| Routing by domain for AI tools | yes | partly | partly | by hand |
| Traffic to AI domains stops when the tunnel drops | yes | partly | depends on setup | by hand |
| Clients update themselves | yes | yes | yes | no |
| Console with devices, usage and audit | yes | depends on plan | yes | no |
Questions
Does it work with Claude Code, Cursor and Codex at the same time?
Yes. Routing is by domain, so the domains of Anthropic, Cursor, OpenAI and JetBrains use the tunnel at once.
Whose servers does it run on?
Yours. The console and the nodes run on Linux servers you own or rent, and the address the AI services see is the address of your server, not a shared address of a VPN provider.
What happens if the tunnel goes down?
Traffic to the listed AI domains stops instead of going out directly, and the client works to bring the tunnel back. Everything else on the machine keeps working as usual.
Do developers have to configure anything?
They install the client and open the profile link the administrator sends them. Updates arrive on their own and are signed, and the operator sees the status of every device.
How do we start?
Send the form below. We agree on a pilot on one of your servers for a few developers, and you decide from there.
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.