> The updates include the addition of memory storage to its developer tools for using AI models. This could theoretically slash costs for application makers by as much as 20-times...

What kind of memory storage would that be? Doesn't make a lot of sense without further details.

Total guesswork on my part, but at the simplest level it could be something like they'll let you store context on their system (for example, in a long system prompt) and let you pull that in more easily than sending it over the wire on every request. At a more complex level, it could be something like having their own RAG for knowledge you upload and to automatically use it to supplement your future requests, but that feels far less likely to me.