There are a lot of databases to choose from and there are a lot of choices about how to connect to them. Whichever database you use, you will need some sort of library to get data to and from your data store. If you are getting to know a database for the first time, then just using a simple client librray makes sense. However, if you want to use a library that maps to the data structures in your code, then you might want to look at using an object-relational mapper. The other alternative is to just use a database with data structures that closely match the data structures in your code (like Redis).
A client library (or driver or adapter) provides a connexion to a database and access to commands specific to that database.
Good examples of client libraries are the Redis client for Node.js or the MongoDB Node.JS Driver.
An object-relational mapper is a single library that allows an application to access a database, often without having to worry too much about which database is actually being used, without having to care about database-specific commands, and without using SQL.
Good examples of ORMs are bookshelf.js, Sequalize, Waterline and Mongoose. Some ORMs, like Waterline work with many different databases and some, like Mongoose, work with just one.
The idea of using a single database-agnostic library that maps nicely to your data models is compelling, but has been known to cause problems with more complex or highly-trafficked applications.