If you do
with pm.Model() as m:
overall_mean = pm.Normal('mu', 0., 100.)
sigma = 10. # <- replace 10. with a distribution or other information
obs = pm.Normal('observation', overall_mean, sigma, observed=data)
The information from your observation data will be aggregated automatically toward overall_mean by Bayes’ theorem (or as some people call it, backward inference).