You can find a basic introduction to utilities here: https://www.sawtoothsoftware.com/support/technical-papers/general-conjoint-analysis/the-basics-of-interpreting-conjoint-utilities.
I do not understand this sentence, so please clarify: 'We want to look at interactions between some variables the sum of the utilities is 0, so it cannot be just avarages?"
In terms of getting a p-value for each level of each attribute, the easiest thing would be to estimate the utilities with a logit model and then look at the t statistics. If you want to keep with HB estimation, you could go into the draws file and for each attribute level, count the proportion of times the utility was above (or below) zero. That proportion would be analogous to a p-value for the utility of that level being positive (or negative, respectively).