automation

The same route for CI runners, AI agents and headless jobs

More and more AI agents run with nobody at the keyboard: in CI, on build servers, on a schedule. They talk to the same AI services as your developers, so they need the same egress address and the same rules.

one route

Why automation needs the VPN too

A developer's Claude Code session goes out through your server. The agent that runs the same tool on a build server at night should not go out from somewhere else. If it does, your company shows up at the AI service from two kinds of address, and the runner is the odd one out.

One route for people and machines keeps the picture simple: the address the AI services see is your server's, whoever is asking. Routing is by domain for Anthropic, Cursor, OpenAI and JetBrains, and everything else on the runner goes direct.

enrolment

Provision by key

People get a profile link. Machines get a key. A runner or a server is enrolled with a provisioning key, which fits into the script that already sets the machine up, so there is no one clicking through an installer.

There is a Linux client, so build servers on Linux can use the same route.

exit codes

Scripts know what happened

check

Verify before the agent starts

The command-line tools return exit codes. A CI step can ask whether the tunnel is up and fail the job early if it is not.

fail closed

No quiet fallback

If the tunnel drops, traffic to the listed domains stops instead of going out directly. A job fails visibly rather than leaking.

repeatable

Same result in every run

Because the result is a code and not a message to read, the check works the same way in a pipeline, a cron job or a shell script.

api

A REST API for the rest

The console has a REST API. Devices and runners can be provisioned with a key, and the command-line tools return exit codes for scripts. Whatever you automate, the console stays the single place where it all shows up.

visibility

Visible in the console

Runners appear in the console as devices, with their status like everyone else's. The administrator sees that a runner is connected, which version of the client it runs and when it was last seen, without a separate tool for machines. A runner that stops reporting is as easy to spot as a laptop that does.

People and machines share one console, one list of devices and one audit log of administrator actions. The administrator does not keep a second set of rules for automation: the domains that are routed and the leak protection are the same for a developer's laptop and for a runner.

faq

Questions

Can a CI runner join without a person?

Yes. A runner or a server is enrolled with a provisioning key, so nobody has to open a profile link by hand.

How does a pipeline know the tunnel is up?

The command-line tools return exit codes. A pipeline step can run one, look at the result and stop the job before an agent starts if the tunnel is not ready.

Does it work on Linux build servers?

Yes. There is a Linux client, so build servers on Linux can use the same route.

Is there an API?

Yes, there is a REST API. Runners and devices can be provisioned with a key.

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.