PostGraphile is formed of three layers.
At the very top is the PostGraphile CLI. This layer is the most user-friendly and is responsible for three things: accepting common options from the command line; spinning up a HTTP server (or a cluster of them, or none at all depending on what options you're using); and mounting the PostGraphile middleware (see next). Most users should start with this layer, many users use it successfully in production, and it's where we recommend you start.
The PostGraphile CLI wraps the PostGraphile middleware ("library"). This middleware is suitable for mounting in Node.js HTTP, Connect, Express or Koa applications. This layer is responsible for receiving, deciphering and validating the GraphQL HTTP request from the user according to the options supplied; configuring a PG client with the relevant settings; and then sending the query on to the GraphQL schema to be resolved. About 70% of PostGraphile users end up using this layer in their applications; reasons to use this over the CLI include the ability to add Express middlewares before PostGraphile (e.g. to perform rate limiting, sessions, authentication and other concerns) and the ability to take greater control over the PostGraphile system.
Deepest down is the PostGraphile GraphQL schema ("schema-only") itself which contains all the types, fields and resolvers. (The schema is constructed dynamically, so cannot be written to disk.) Most users will never use this level.
The deeper you go in the stack, the more complex your setup code will be, but the more powerful your integration can be. We're always trying to expose as much power as is reasonable through the CLI, but it's not sensible to make every possible thing a CLI option - if you need that level of customisation then you should opt for the middleware.
We recommend that you start with the PostGraphile CLI and then move to the PostGraphile library if you need deeper integration with Node.js.
Which layer would you like to read more about?