Routing by domain, with lists that keep up with the AI tools
AI tools add and change the domains they use, so a list written once goes stale. Here the lists are kept current and the administrator stays in control of every change.
Only the AI traffic goes through the tunnel
Routing is by domain. Traffic to Anthropic, Cursor, OpenAI and JetBrains goes through your node. Video calls, package downloads and the rest of the web go direct, so developers do not notice a slower connection.
Three sources for one list
Your own entries
The administrator adds domains and networks: internal services, a partner API, anything the team needs in the tunnel.
Public lists
Public sources are refreshed every hour, so the list follows the services without anyone editing it.
AI suggestions
The console proposes additions that look like they belong to a tool. They wait for a decision.
AI suggests, an administrator decides
The console asks an AI model which domains belong to a tool. What comes back is a candidate, nothing more. It appears in the console, and an administrator approves or rejects it. Only approved entries reach the devices, and a rejected one stays rejected, so the same suggestion does not come back.
The result is a list that follows fast-moving tools without handing control of your routing to anyone. Every change an administrator makes is in the audit log.
Why a static list is not enough
An AI coding tool is not one hostname. It is an API, a login service, an update endpoint, a telemetry address and a few more, and the set changes whenever the vendor ships a new feature. A list copied into a config file in spring is wrong by autumn, and the symptom is a tool that half works: the chat loads, the agent stalls.
The opposite approach, sending everything through the tunnel, costs bandwidth on your node and slows the developer’s ordinary browsing. Lists by domain keep the tunnel small and the routing explainable: you can read the list and say why each entry is there.
What happens on the device
The client receives the lists signed and checks them before it uses them. When the tunnel is down, traffic to the listed domains fails closed: it stops instead of going out directly. Everything else on the machine keeps working as usual.
What an administrator does day to day
Most of the time, nothing. The public lists refresh on their own every hour. When a suggestion arrives, the administrator opens the console, reads the domain and the reason it was proposed, and approves or rejects it. A team with its own internal services adds them once, by hand, and they stay.
Each decision is recorded in the audit log, so you can later answer who added a domain and when.
Questions
Can we add our own internal services to the tunnel?
Yes. The administrator adds domains and networks by hand, next to the lists that ship with the product.
How quickly does a new domain reach the devices?
Public sources are refreshed every hour. Entries added by hand and suggestions an administrator has approved go out with the next list.
Does the AI change our routing on its own?
No. A suggestion from the AI model is only a candidate. It takes effect after an administrator approves it.
Which AI tools are covered out of the box?
There are lists for Anthropic, Cursor, OpenAI and JetBrains. Anything else you add by hand.
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.