The interface is protected by Django authentication: you can execute one or more SQL queries, copy the URL and share it with your collaborators on that project.
You can also create a saved dashboard and make that available to the wider world. Here are a couple of examples of saved dashboards that I'm running on my personal blog:
These types of tools are extremely handy.
That said, part of my rationale for building this was as a learning and research exercise since I need to learn more about read-only PostgreSQL for my Datasette project.
Definitely looking forward to cross-pollinating ideas. I agree, having access to this kind of interface is a massive productivity boost.
I've had a lot of success with http://redash.io, which is a similar (though much more complex) project. SQL backend +visualizations can be really powerful and surprisingly flexible for building dashboards that aren't quite ad-hoc, but don't deserve their own standalone app and you just need to tie some input fields into SQL parameters.
edit: Added Djangonaut mention
Read replicas are pretty easy to set up with most managed database products (Cloud SQL, AWS RDS), so that would be a very lightweight way to add some ad-hoc querying capabilities in prod that won't keep you up at night.
The key feature here is that you can build a new report without having to write any Python/Django code at all.
I know Jupyter exists - but a solution like this with the permissions would be valuable.
pip install django-extensions
python manage.py runserver_plus
This should run IPython Notebook with database models already imported :
python manage.py shell_plus --notebook
The buttons also reminded me of Goat Counter which also employs the minimal CSS things