We want to emphasize how important privacy is to us. It felt wrong to be able to access a company's private knowledge so we set out to make the system as secure as possible. In terms of our embedded vector DB, we use Homomorphic Encryption to vectorize the embeddings. We are still super early - started hacking on this project 7 weeks ago, in the near future we plan to get SOC 2 compliant so we can definitively prove that we are secure.
You can read more about Homomorphic Encryption here: https://aclanthology.org/2022.finnlp-1.4.pdf