Obs30 I could run this proxy maybe on an iPad.
I wouldn't place an iPad out in sub-freezing weather though.
A cheap Arduino should do the job (using either C, or CircuitPython). The Arduino Due, for example, has two "external" RxD-TxD pairs that you can connect to two FTDI modems. One side then plugs into the real mount, and the other FTDI modem looking like a different command protocol, that plugs into the computer.
The latency can be in the tens of milliseconds (but not too much more) and still handle pulse auto-guiding well. Recall that at 9600 baud, the data rate is about one character every millisecond -- and a slew command to do pulse guiding will typically take 8 to 10 milliseconds at that baud rate. The proxy will double that latency, so you may have a little to spare. But 115,200 baud should not be a trouble at all.
I don't use Swift, by the way, but Objective C. I was using Objective-C at Apple even before MaxOS X was released to the public. I retired back in 2005, which was way before Swift became a project. I'm still constantly writing programs (a hobby since 1966, when I encountered my first computer), most recently writing a program for simulating point spread functions (kind of like the Maskulator, but with ability to apply the quadratic factor in Fourier Optics to also simulate out of focus point spread functions and MTF, and Wiener filters to correct MTF, etc. I can post that Xcode project to my Hostgator server if you are interested -- it uses the native Accelerate Framework to perform the 2D FFT, which could be useful sample code, too.
By the way, simulating the electronic focusers are more problematical, since ZWO uses USB-HID to control the EAF, and they do not publish the protocol. But that one should be easy to solve by using a "mechanical proxy." I.e., a real EAF connected to a shaft encoder, and the shaft encoder is polled by a micro-computer that controls a real focuser. One day, I might try that trick to use my FocusLynx with the ASIAIR. It might be easier to hack an EAF by opening one up and polling its stepper motor lines directly instead of using a shaft encoder -- that way, you won't have to contend with the large backlash of the cheap EAF stepper motor, and end up with the native (better) backlash from the non-ZWO focuser.