I've been experimenting with InfluxDB a bit lately, and it is a very cool product. It allows you to store any sort of timeseries data, which is obviously useful for what we are trying to do here.
It's main interface is over HTTP (though there is a python client, making ad-hoc queries at various points in your algorithm wouldn't be very performant). Quantopian does provide the fetch_csv function, though, to load CSV data over HTTP. The problem is that that InfluxDB doesn't output CSV.
So, I wrote a small proxy app that Quantopian can hit, to repackage the query results as CSV. I have set it up on a free Heroku instance for trying it out, but it is open source if you want to inspect it or host it yourself.
Hope someone finds this useful!