Fiber Optics

For short range (less than 10 or 12 feet), you can use poor mans fiber optic cable. It sounds bad at first, but it DOES WORK, and the best part is that you don't need special RF chips, antennas, it's interference free and can be contained in the existing enclosures. It's also RUGGED as Hell, easily concealed and cheap!
Transmitters are simple LED's and receivers are photodiodes. The fiber optic cable is 20 to 40 pound test fish line, make sure you get the clear kind and make the ends with a razor sharp blade in a singe cut.
I am currently using this for communication between my keyer and my ham rig. This Fall, I hope to convert my wall mounted t-stat to a similar system to avoid an rfi problem when we transmit on the ham radio at high power.
If you need more range or a higher data rate, use a laser diode, find the focal point of the collimating optics and put your fiber optic cable there. I don't know the range of this setup, but would guess it to be 1km for narrowband and 100 feet or more for medium bandwidth. I HAVE NOT tried the laser transmitter myself, but the sub mini surface mount (RED) LED's work like a champ.
Snag an LED out of a late model piece of consumer electronic scrap gear and look for the sub mini smt type led's (the kind with the flat leads coming out of the sides of the package 180 degrees apart). While it's still mounted on the pcb, file down till you almost hit the die--making a flat face on the front of the led package. Now, use some Comet or abrasive containing toothpaste and polish the front surface on a piece of glass. When the LED has a flat and smooth face, you can attach the fiber at point blank range. I like to attach the cable to a small chunk of plastic with a pin hole in it, the plastic needs to be .25 inches thick and snug when the line is forced into the hole. The plastic is the only thing that keeps the cable at right angles to the LED's active area. Glue the led so that it shines through the pin hole. After the glue sets, you can slide the fiber optic cable into the hole until it butts up against the flat surface of the LED. After the assy is glued and tested, be sure to coat the entire assy with something that doesn't let stray light in (any stray light that gets into either end of the cable is interference and degrages the signal to noise ratio).
Before hanging the messenger or having him committed, try sighting down a piece of fish line with the the far end close to a light and you will easily observe the light transmission capability of the poor mans fiber optic cable.

Fr. Tom McGhee says:

You might try simple acrylic rod. Cut the rod, use fine sandpaper to get the ends relatively smooth, then dip the ends in the solvent used to join acrylic. I believe it is methylene chloride (but could be wrong). Anyway, let ends dry and you will have a shiny clear flat end.
The acrylic rod can be heated and bent if desired. Finish the ends AFTER bending, as the bending can slightly deform the rod ends.

A great fiber optic demo kit was available from Mouser electronics. They may have discontinued them

DIGIKEY has :

FB104 $10 (simple--digital link, 1 meter or so--optic parts only No PCB)
FB105 $17 (complete kit, all parts including PCB, you assemble)

Digikey also has course kits, and other stuff up into the $100's.

The parts in any of these kits were made by Motorola, and have been out of production for many years.

The same parts are made now by SIEMENS; the SFH756 and SFH551 transmitter / receiver pairs.

Peter L Peres says:

I have gone to 38400 Bps using lasers and detectors stripped from CD player laser units. The receive amp was a venerable uA733 used at max gain (x100) followed by a 393 comparator. Optics were 1/2 binocular (3x10 theatre type I think). The far side was a catadiopter affixed at some 60 feet. Both Tx and Rx used binocular halves (of the same unmodified binocular). The binocular I used had UNCOATED lenses and no IR filter. Obviously.
Interface to PC using RS232 and homebrew test program (direct register access serial driver etc written in C). The test program did bit stuffing (detector threshold was automatic AC type) and made sure that the laser was on at most 1/2 of the time. The lasers I tried were 60 mA dc so I ran them at ~250 mA peak (with ~50% silence and 50% max duty cycle). I think that I achieved some 10E-4 error rate after serious tinkering, and then I gave up. Pointing a IR laser manually without helper devices across 60 meters is not something you want to do in a profession you make your living out of imho. It used to work best in absolute darkness (night + any streetlamps removed from the FOV).
Actually I know of one ancient project that used a headphone speaker (bared, dome painted with silver paint), to act as a light modulator and send voice across great (100 metres, using telescope optics) distances. The light was halogen 12V 20W, with mostly IR used (filters). I do not see why a laser could not be used in the same way, with direct AM modulation. Receiver was ac coupled LDR + LM386 style amp and phones.