End-to-end encryption

[morajabi]

Hello all,

This is a discussion to gather feedback and ideas from community and experts on end-to-end encryption in Inline.

We'd love to give the users and ourselves peace of mind but there are technical challenges and trade-offs that resulted in not implementing it at this stage.
Currently due to technical limitations, conflict with other core features we have in mind, and our very small team, end-to-end encryption is a non-goal in Inline. If any of the three aforementioned matters change, we'll re-consider.

These are my notes on some of the limitations/challenges:

• Bot DX: Implementing bots must be really easy for the community. Is it possible to prevent E2EE from over-complicating this?
• Sync: From a UX standpoint, cross-device syncing in Signal seems to be noticable slower than cloud sync in Telegram for example. Is there any research/paper on how to make this fast?
• Large communities: Would we need two types of channels so for larger/public communities we switch to a cloud-based method?
• AI: More and more users expect their work apps to support AI features such as search, completion, processing, classification, etc. How would this work? Especially for RAG/generating embeddings or captions for images which need to be passively processing messages. Would those users would have to just disable end-to-end encryption? I expect majority of the users wanting such features which would make E2EE mostly pointless at least in channels.
• Media and files: This is more a question. I don't have an idea how this could work for a workspace seamlessly while providing E2EE.
• Search: Local search is great, and we're implementing it. But users coming from Slack expect instant search from all their devices across the space. From what I understand, we'd have to download the whole space data in the background to enable this.

Alternatives

• Only in DMs: This could be a balance between privacy and user experience. But still doesn't solve the whole trust problem.
• Self-hostable: If the above concerns become impossible to solve, we could explore making a self-hostable open-source version available instead.