Thanks for the two(!) nice tips. Just looked at libcaca, but I'm having trouble installing on my laptop (OSX). Anyway the idea is mainly to remove gnuplot as a dependency from my scientific code; introducing a new dependency defeats the point a bit.

Jupyter would be a very nice way to interact - as you guess there are difficulties: playing well with the cluster queueing system, and currently the code is launched via an mpirun call. I'm open to suggestions.

Edit: This sounds like exactly what I need! http://ipython.org/ipython-doc/stable/parallel/parallel_intr...