Small clarification about the demos.

The goal of the prototype is not to build “another VPN”, but to experiment with a different connection model where the session identity is the stable anchor and transports are replaceable.

In the demos the session survives events like:

• transport failure • relay change • path replacement • NAT rebinding

The runtime treats these events as state transitions rather than connection resets.

Current prototype includes:

• session runtime with deterministic state machine • transport abstraction layer • relay forwarding experiments • trace-based invariant verification

Right now everything is still very experimental.

The main question I'm exploring is whether a session-centric model could simplify recovery logic under network volatility compared to tunnel-centric designs.

Hi HN,

Building a VPN where session is the stable anchor, transport is volatile/replaceable.

Pushed new 60-second demos: - Multipath failover (one path dies → seamless switch) - Path kill & recover - Distributed ownership takeover - NAT rebinding test

Also: basic UDP echo server runs in Termux on Android phone.

Repo: https://github.com/Endless33/jumping-vpn-preview

Short video (UDP test): https://vm.tiktok.com/ZNRuY8uug/

Looking for feedback on: - session vs tunnel-centric approach - invariants / state machine - next: QUIC + real multipath

What do networking / Go people think?

Another small detail: the prototype emits structured trace logs for every session event.

This makes it possible to verify protocol invariants and replay execution traces to debug recovery behavior.