My guess is they are using Embeds to efficiently search a larger corpus of data and picking some data sources that then get injected into the system prompt.

That's how I'm doing it for myself and how many other companies are doing it to enable doc interaction, etc.

Edit: Wish I knew the term RAG (comment next to me mentions it) before going down a rabbit hole of trial and error with the limited amount of info on Embeds

I found this to be useful https://aws.amazon.com/what-is/retrieval-augmented-generatio...

They also allow for creating custom actions using structured data to external APIs and file upload for RAG type use cases. And they can call out to code interpreter, DALLE, and Bing