Thanks for the question Michael. Currently, the Knowledge Table repo ends at the CSV/triple creation / export stage. You can define the CSV columns in the platform itself, or in the triples that are extracted.

However, I assume you are thinking of how that mapping is stored and retrieved for querying. The querying and memory aspects are part of the WhyHow platform which is closed source, and will be part of the hosted version we launch.

Regarding your second question - basically yes-ish. That's definitely one design pattern we advocate for. But you can also imagine that an end to end process where the query is sent to a system with no pre-existing graph. The query is run against multiple documents, the answer is generated, triples are generated, stored in a graph and then queried against.