Phabricator isn't code hosting, it's repository viewing. The code is still hosted in whatever existing SCM tool is being used. It does have a lot of features that overlap with github which is why it is a bit confusing.

Phabricator does have code hosting, and the Phabricator master repository is self-hosted on Phabricator and mirrored to GitHub.

We link to GitHub because they have more infrastructure than we do for handling heavy read loads.

How is the code hosting done in the backend, do you reuse git or is there something bespoke in action back there?

Underlying writes go to Git, Mercurial and SVN.

Generally speaking, we proxy requests and pass them through to the VCS after performing authentication and broad permission checks. Requests are rewritten slightly to remap user-facing display paths to actual paths on disk. Some additional features act through commit hooks.