Local-first promise vs optional server acceleration

[Abhash-Chakraborty]

Context

Find should stay local-first. Users should be able to upload, index, search, and browse their private images without sending them to someone else by default.

At the same time, heavy ML models make installers large and mobile support difficult. It is reasonable to discuss optional server acceleration, but only if the privacy boundary is explicit.

Non-negotiables

• Default mode stays local.
• No silent upload of user images or embeddings to a hosted server.
• Any remote mode must be opt-in and clearly labeled.
• Users should be able to point to their own machine or self-hosted server.
• The app should explain what data leaves the device before enabling remote processing.

Possible optional modes

1. Self-hosted accelerator

   • User runs Find backend/worker on their own desktop, NAS, or server.
   • Mobile/desktop clients connect to that user-owned endpoint.

2. Bring-your-own model endpoint

   • Advanced users configure a local LAN or private server endpoint for captioning/embedding.

3. Hosted demo mode

   • Useful for demos, but risky for the project identity.
   • Should not be the default contributor direction unless privacy terms and scope are very clear.

Discussion questions

• Is a self-hosted accelerator compatible with the local-first motto? I think yes, if the user owns it.
• Is a project-hosted server compatible? Maybe only as an optional demo, not the main product.
• What should the UI say when remote processing is enabled?
• Should remote mode be blocked for private libraries unless users explicitly opt in per library?