Does PyMC3's Timeseries API allow for time varying parameters of any model that would require fixed values under a frequentist approach?

Before diving into PyMC3, I would like to know if it can solve my problem.

I’m dealing with time series modeling and my problem is that my time series exhibit frequent structural breaks that make every inference a nightmare. Usually, my final goal is to bootstrap the models’ residuals and perform some simulations, but this often turns out to give inconsistent results if I’m not able to drop the sample before a structural break.

Instead of relying on the classical approach for detecting structural breaks (Chow, CUSUM, and so forth), I would like to try something a bit different.

According to examples like this, PyMC3 allows not only for the usual Bayesian prior/posterior distribution estimates but also for these time-series models. It looks like that any process defined for example as a GaussianRandomWalk will return a time-varying estimate. See the following example where a GaussianRandomWalk has been used:

The only difference with my desired setup is that here above the output of the case study is the GaussianRandomWalk itself, while I need something that resembles a latent variable. For example, I would like to model the linear regression slope between two time-series by using a PyMC3 time-series process, like AR1 or GaussianRandomWalk, and walk away with my estimate of $beta(t)$ like they did in the picture above with $s_i$.

This way I could face the problem of the structural break by taking only the latest parameters values because they’ve been modeled as functions of time.

Question: is that something I can do with PyMC3 GaussianRandomWalk and the other time-series API? Does this machinery applied to any parametric model return the time-varying parameters I’m looking for?

If you have any different R o Python package that you would like to suggest to address this problem, feel free to suggest!