I've been prototyping simple desktop GUI tools on top of dask/pandas and PyQt that let you lazily load large CSVs (and other types supported by pandas) and interactively filter based on smart histograms (the per column histograms are fully interactive and provide crossfiltering across the attributes):
The idea is to map a lot of the basic functionality of dataframes onto simple GUI interactions (for example, changing column types, stacking and unstacking columns, pivoting) and couple that with an ipython console for more complicated data manipulation. And then maybe even adding adding Tableau like charting functionality:
For quick throwaway exploration/analysis. It can easily handle about a million rows just using generic pandas and a bit of memory. There's lots of cool database techniques that can also be used on small local data (for example, compressed bitmaps using EWAHBool for interactive filtering).