It says that it only stores metadata locally not objects themselves (although it could probably have an option for that too and serve as a general S3 cache).

Nice effort, Redis is perfectly fine but I believe that the storage layer should be somehow more separated in case someone wants another type of storage, e.g. in-memory SQLite is adequate and already installed in most systems.

Yes, I think that's the gist of it. It looks like it assumes it has seen every call to the bucket it's working on though. So to start using it with an existing bucket, you'd have to warm up the Redis cache by downloading all metadata from the bucket?

Exactly. If you're using boto.s3.connection or boto.s3.bucket you can change that to mimicdb.s3.connection and mimicdb.s3.bucket.

I think it is a proxy for S3 that sits between your application and S3. So it works with Boto, or your favorite S3 library.