The data is quite simple. At the most basic level it's just a photo of bright spots on a chip, as read by the machine. Knowing that spot A corresponds to variant A and spot B to variant B, an algorithm then decides ("calls"), depending on the relative brightness whether the person has variants AA, AB or BB (or impossible to tell). This is the only real processing and there IS open source software for that (packages for R, most famously CRLMM).
So, the whole point is getting the variant calls, ie what sort of nucleotide the person has at specific positions, for example rs314159 --- yes I chose that based on pi, but it does exist. If you have the variant calls, you can then try to decide, based on available literature, what this means for your health.
The not-so-obvious reason why SNP genotyping has not yet made it to the clinic is that polymorphisms (the stuff that genotyping arrays discover) are either frequent or associated with significant effect, but rarely both. Some rare variants (for example BRCA1/2) have important consequences and some variants are very frequent (for example, for hair color) but don't have important health consequences. This is the result of natural selection: bad genetics tend to get thrown away and become rare.
Simply put, most of the information associated with frequent polymorphisms only modifies risk by a relatively weak amount (relative risk 1.2-2, for example) and may also depend on other polymorphisms or entire haplotypes (a whole bloc of DNA) or even the environment. The resulting information is NOT of sufficient quality to dictate anything beyond things that we already know, ie don't smoke, eat healthy, moderate exercise etc. There are a few exceptions, for example in pharmacogenetics, where some people react differently to drugs, especially important or highly toxic drugs (clopidogrel, 5-fluorouracil, irinotecan etc). In these cases, there is some interest in genotyping and the FDA does mention cases where it could matter. Nevertheless, genetic testing for pharmacogenetics is not universally performed and is not generally required.
In the end, at the current state of affairs, the information provided by 23andme is most useful for ancestry but not particularly useful for health-related decisions. Which is why the FDA stopped them in the past.
Anyway, don't underestimate the interest 23andme has in farming your data (like Google). Even anonymized data without phenotype correlations can be scientifically very interesting. This is not necessarily a problem, but is likely to be the case with all similar genotyping offers.