HTM has been implemented in many repos. The original ones were created and maintained by Numenta, called NuPIC and NuPIC.Core. These are now in maintenance mode. They were forked by the community into the primary htm.core repo. This is now the most active HTM repository I know of, with several experienced forum members contributing.
There are many other implementations of HTM in different languages and environments.
Numenta is currently working on applications of the biologically-inspired ideas behind HTM to today’s machine learning frameworks. See How Can We Be So Dense? The Benefits of Using Highly Sparse Representations. You can see this in nupic.torch and (work in progress) nupic.tensorflow. These repos do not contain complete HTM implementations, but aim to apply ideas of sparsity & robustness inspired by HTM into these platforms. You should use these if you are building deep learning networks and want to try applying HTM there.
If you are just trying to understand HTM and the biologically-inspired ideas therein, I suggest you implement HTM yourself. The best resources for this are BAMI and Building HTM Systems, both of which are still incomplete. However, BAMI contains all the pseudocode you would need to implement Spatial Pooling and Temporal Memory.
I know it is confusing to go to the Numenta github and see so many old repositories. I am going to work on archiving those we are no longer user or promoting.