very basic API-keys - #300
Conversation
|
@erik-elmgren Thanks for PR. There are some issues with the PR, as it breaks existing pipelines that use the audio.cpp server and the implementation is incomplete. However, my biggest concern is that it goes against the current design principles of the server. In general, the server is intended to stay lightweight. Features that require additional application-level state or infrastructure are generally better handled by a layer in front of audio.cpp (e.g., a reverse proxy) rather than being built into the inference server itself. You can check the discussion here for more context: #173 That said, I’m happy to reopen the PR if authentication is implemented as a complete, self-contained, and fully opt-in subsystem that does not affect existing server behavior or pipelines when disabled. And please don’t let this discourage you from contributing! |
Let me know if this is interesting.
I think it would disable the web gui at the moment as there is no easy way for a browser to add a authorization bearer header to all the requests.
Obviously https is a next nice thing to have to make it fully useful.