From a pure contention device we extended the TAC to give a choice: access to The Mainframe or to Somewhere Else (LSE was one, IIRC, connected to us, as with all external connections, by various muxes, statmuxes and who-knows-what ad-hoc links). So we had a sort of network device.
After a while a couple of wizzkids joined and built stuff with Z80s in them: one device was an interface to our 'KingMatic' - a humungous flat-bed plotter controlled by a triple-bay enclosure full of PC boards of discrete transistor circuits, each having the function of one 7400-series IC!. Another was a colour-hardcopy device: this was built out of a 35mm camera with 200mm lens in a darkened room (blackouts over the windows) pointing at a high-resolution monochrome monitor about 8 feet away. A frame-buffer (built on a wire-wrap circuit board) connected to a Z80 micro (on a bought-in board, the 'OEM-80', with CTC, SIO, PIOs etc) which generated various planes of a 3-colour image which were displayed for various periods of time (t, 2*t, 4*t etc) on the monitor, whilst an incredibly Heath-Robinson affair of toy electric motors, fishing-line and pulleys on a 'Meccano'-like frame moved red, green and blue filters in turn in front of the camera lens, whilst the shutter was held open by a cable-release operated by an old radio-control-model servo donated (by me!) to this hi-tech project. Motors and servo were of course also controlled by the Z80 system, and the camera had a power-winder which advanced the film after each exposure, so the system could be left running to generate pictures entirely under control of the computer. Images were usually generated on the mainframe, and sent to the system over a serial link.
But I digress ... :-)
The wizzkids then built a Z80-based PAD with a bunch of 16 or 32 (though the 32 didn't really work too well) serial ports to connect to terminals, and a high-speed (380K-ish) network connection (using RS422 with self-clocking Manchester encoding (IIRC) over just 2 pairs of telephone-grade cable) with HDLC signalling and a sort of stripped down X.25 with dynamic addressing connecting to a switch (same device, sans V.24 connections) in the core of the network. This was at the time that most academic sites were using 'Camtec' PADs which had 16 serial ports (capable of running up to 9600 baud) and a 19,200 baud network link. These required an expensive X.25 switch in the core, modems or line drivers to run PAD<->switch links over 2- or 4-wire circuits, and network addresses had all to be set up by hand on each device. Ours were cheaper, faster and easier to manage (with one exception ...)
One 'feature' (as we'd call it these days ;-) was the watchdog timer: if the processor on the CPU board hung, the watchdog timer should reset it. Unfortunately a logic error (or 'thinko') got this reversed in the PCB design, so if the CPU was running it would reset itself ... this was never fixed, so for years afterwards a frequent chore in Network Support was to go round resetting hung CPUs on the switches, and chasing round resetting hung PADs in departments (most of which were in locked store-rooms with inoperative lightbulbs, 7 feet up on a shelf with no nearby access :-).
As we went from making one-off and prototype devices to producing, eventually, several hundred PADs and many switches (plus gateway devices), the production of this kit had to be put on a proper footing. The College did not want to become a manufacturer itself, and was unhappy to buy thousands of pounds' worth of networking equipment from a bunch of its employees moonlighting in their spare time, so a Company was formed: Netcomm Ltd. (This was still a bunch of employees moonlighting in their spare time, but the College had a nice feeling that it was dealing with something solid and respectable!)
After the local (college) demand for equipment was met the wizzkids who were Netcomm developed the X.25-like product into real X.25 kit: switches, PADs and a management station (called, IIRC, the NMS2) which they sold to many universities and other organisations, and went on to develop equipment using the then-new technology of putting everything in 48-byte cells carrying Data, Voice and Video etc. Their management station developed into the NMS3000: a Unix-hosted, Motif-GUI, TCP/IP-based system along the lines of HPOV Network Node Manager (though they claimed theirs to be much better than OV :-).
Somewhere along the way they got bought up by a US company General DataComm, out of which the wizzkids (who were shareholders in Netcomm) did very nicely, but since then they seem to have gradually disappeared from view (at least, the view from where I sit).