A pty is owned by a process. When that process ends, the pseudo-terminal is torn down and everything running in it receives a hangup.
If the owner is the application window, then reloading the interface kills every session in it. This is why most embedded terminals lose your work when the page refreshes, and it is not a bug in their implementation. It is a consequence of where they put ownership.
What the bridge does
It holds the master end of every PTY, keeps a ring buffer of what each one has printed, and speaks to the interface over a socket. The interface becomes a view: it attaches, gets the buffer, and streams from there. Detaching changes nothing about the session.
That single move buys reloading the app without losing sessions, restarting the server without losing sessions, and attaching from a second device.
The isolation that stops being optional
A bridge is a long-lived process spawning shells, so anything ambient it inherits is inherited by every session it ever starts. Which means a test run, a dev instance and a production instance must not share a data directory, a home directory or a socket path, or a probe against one will terminate the sessions of another.
That failure is worth naming precisely because it looks like something else: to the person whose terminals just died, it reads as a crash, and the actual cause was a health check somewhere else on the machine.
A separate bridge process owns the PTYs. Reloading the app or restarting the server preserves the sessions and their scrollback.