Well, a database dump is fine if you have some new metadata to manage it. Basically, a versioned schema ... Imagine that you end up wanting to synchronize you

Well, a database dump is fine if you have some new metadata to manage it. Basically, a versioned schema ... Imagine that you end up wanting to synchronize your DB dump with someone else's DB dump. The thing that's nice about the API is that you can be loosely coupled with it. Having access to all of the data is both a blessing and a curse, because you can't escape a stronger coupling. Or, if it's couchdb, then each object has its own revision. The parent blog post is anticipated by one of the major use cases of couchdb in the "millions of couches" vision. Having a nice RESTful API is great for some use cases, but consider that you're on a high-latency connection. Here, it would be much better to use a local, embedded database instead of the REST interface. Take a look at couchdb - this is already possible.