I like it. Does this perhaps do the same thing for you as running a caching pip would? Do you think it's still worth adding a feature, or an example, for using a custom pip server?

I think it's enough with this. I mean, I don't see a real difference at this point between this simple option and configuring a devpi for example. devpi is good when you have multiple users on the same network but for development... This is enough and very simple to set up.

👍 I'll do some testing with this setting too

I just noticed that this is not enought to make it work properly.

You can set GLOBAL_PIP_CACHE to whatever you want and everything will continue working but no package will be stored permanently in the disk. Since the command runs inside the container we need to set a path mounted inside the container to make the downloads permanents.

Summarizing, we can just write data permanently only on the directories mounted in the docker container: https://github.com/rtfd/readthedocs.org/blob/2c946d15cb2a43425ff829120e5fc75c19c5a494/readthedocs/doc_builder/environments.py#L639-L652

None of those are good because some are template path and the other is very specific for the current project being built. So, the solution would be mount the self.project.pip_cache_path as well. Like this:

                    self.project.pip_cache_path: {
                        'bind': self.project.pip_cache_path,
                        'mode': 'rw'
                    }