Sounds quite interesting, a database with pseudo-transparent access from Python. Objects are automatically read in and automatically written when changes occur. Objects are automatically added to the database when they are referenced by an already stored object. Databases are cleaned of garbage via reference-counting (unless you produce circular references). And there's checkpointing, which allows you to ensure that a database restarts with the last consistent state. In a broad sense similar to Metakit, but somewhat more focused on objects than on tables. This certainly allows for more elegant integration into Python code. The question is, how is the performance? Because many small object databases are terribly slow when the number of objects in them grows. And large object databases are simply overkill for something like a weblog tool. Here's the original article.