Multivariate random walk parameter

I have a parameter alpha that is time variant. It has a transition matrix T (parameter) and also depends on another covariate x2. In stan I would have a loop

alpha[, t] ~ multi_normal(T * alpha[, t - 1] + b[2] * x2[t], quad_form_diag(Qc, Qtau));

How could I write this is pymc?