Both in terms of the models performance and in terms of it’s security architecture. Seems to be the most secure remote AI available today, much more so than Lumo, as it runs in a TEE.
Is this something you would trust with sensitive information?
What’s a TEE?
It’s a trusted execution environment.
This means your query, though plain text, is processed in a hardware isolated environment which means it should be impractical to exfiltrate. I think confer.to has some cryptographic handshake it does to ensure this hardware separation is taking place before every query is sent off. The chat history is then e2ee, but that’s the easy part. It’s the processing where the difficulty lies.