I'm bootstrapping a privacy respecting, multi-model AI chat service

ChatGPT and friends have made generative AI easy to use however data privacy is an afterthought with companies openly saying that conversations will be used for training and can be reviewed by staff.

Self-hosting can offer privacy but requires technical knowledge, adequate compute power and often comes at the sacrifice of nice cloud ux features (e.g. syncing across devices).

I'd like to build a privacy first option[0], bringing zero-access encryption to users data to offer some reassurances that you are in control of who has access to your chats. This has been a successful pattern for email with ProtonMail, Tuta and Skiff (RIP) taking email plain text and encrypting it for their users.

In the first iteration I'm aiming for the following features:

    - Encrypted core. Messages and generated responses currently have to be transferred in plain text however they should only live in short-term memory and be encrypted as soon as possible.
    - AI Chat. It's simple, we're familiar with it and it's useful.
    - Multi-model. Similar to Quora's Poe, I'd like to offer the ability to use different models both commercial and open source.
    - Conversation sharing and collaboration. Choose to securely share your encrypted chats with other users.

In the spirit of Steph Ango's recent post[1] I am aiming to be 100% user supported to remain aligned on delivering privacy AI tools rather than pressured to take actions to please investors.

I'm fortunate enough to have ~6 months time and runway to bootstrap building this however any help to extend this would be much appreciated. That can be financial or can also be simply sharing and signing up to receive email updates.

Am I the only one who is looking for a more privacy friendly AI chat?

Happy hacking, Ewan

    [0] https://cognos.io/
    [1] https://stephango.com/vcware