/how-it-worksHow it works
How per-resource routing works
What happens between pressing Connect and Claude returning its first line of code.
01Four route types
Why each route exists
No detour where none is needed. Decided at the application layer, targets that need no optimization stay on your own network and use no accelerated traffic.
The bottleneck is usually the long-distance stretch over the public internet. Global nearshore edge nodes replace that stretch with an optimized line: lower latency, less packet loss.
Many AI services care about the quality of the source IP. Entering through an edge node and leaving through a compliant IDC exit balances speed and availability.
A few AI tools, Claude Code above all, are extremely strict about IP type and cleanliness, and direct residential lines tend to be slow. So traffic is accelerated nearby first, then chained to a dedicated residential exit.
02Separated nodes
Edge nodes and exit nodes each do one job
Either kind can be scaled or replaced on its own without touching the other, so you never have to choose between fast and usable.
- Deployed in regions close to you
- Optimize the long-distance link
- Accelerate only, never act as an exit
- IDC exits and residential exits
- Keep services reachable and lower risk checks
- IP sharing limited by plan
03Unified auth
Unified authentication and access leases
- Every device has its own credentials. The private key is generated locally and the server keeps only the public key.
- Nodes keep no user list. Instead they sync access leases from the control plane: who may use which routes, and until when.
- Bans, password changes and device removals sync to every node in real time through an event stream, so revocation is immediate.
- Standard and Pro users are not tied to fixed nodes and still get globally consistent auth and auditing.
04No silent fallback
We pause rather than switch quietly
When the required exit is down, we tell you plainly that the resource is paused and why, instead of sending it out another exit. AI platforms usually treat a sudden change in an account's exit IP as a warning sign. Traffic resumes automatically once the exit is back.
05Specs
Technical specifications
| Tunnel protocol | Carried over standard SSH; nodes run our own user-level SSH service, isolated from the system admin channel |
|---|---|
| Local proxy | HTTP / SOCKS5 on one port, listening on loopback only |
| Connection modes | PAC / explicit proxy / TUN |
| Route policy | Signed by the server, decided per resource and plan |
| Network adaptation | Re-probes the base network and re-routes after changes; never bypasses corporate security kill switches |
| Device credentials | A separate key per device, no private keys on the server; rotation and revocation supported |
| Access sync | Access leases plus a real-time event stream, acknowledged after nodes persist |
| Client updates | Signed release manifest, health check, full rollback on failure |
| Platforms | Windows x64 / macOS (Apple silicon, Intel) / Linux x86_64 |
Ready to stop waiting on the network?
Sign up in a minute. Connect in one click.