Allow users to opt out of certain ai providers

Thibaultmol

If I as the user don't trust certain ai providers, it would be nice to explicitly disable those providers.
This is definitely relevant with the current openai situation where users might want to disable their query from going to openai's api.

Currently Azure's openai service is the main providers for openai models with open ai's api as the fallback. But I feel like there should be a setting that allows me to disable the openai api.

The way I see this working is that if openai is disabled and I do a prompt to an openai model. it'll try azure first, if that fails it should ask if I want to

Retry with current provider
Send prompt to [disabled provider]
with a [ ] remember this for the rest of the thread, checkbox

Alvir

As a note to this, it would be nice to make sure that Kagi is using any and all "zero data retention" options where possible as this would get ahead of any similar issues (in theory)

(I would still like this option, but just to get ahead of things, so that we don't end up trusting a provider for a while only to have our data shared after the fact)

Thibaultmol

Alvir forgot to mention it here: So I updated the docs recently. Kagi indeed doesn't have a ZDR contract, see docs foonote:

https://help.kagi.com/kagi/ai/llms-privacy.html

¹ When our Azure OpenAI Service API quota is saturated, the request is sent to the fallback provider: OpenAI's own API. OpenAI is currently required by court order to retain all chat logs. We are working with Microsoft to raise the Azure quota and minimise these fallbacks.

utopia

I strongly doubt that Kagi would qualify for ZDR for OpenAI (or similar). While I don’t know OpenAI’s policies for giving an organization access to this feature, we can refer to what Google asks for to opt-out of abuse monitoring (ZDR basically).

Google asks for:

How you use AI (generating or summarizing);
Who can access your AI (internally or regular users);
Child safety protections in place;
What sensitive industry are you in that requires opting out;
Is there someone that monitors & acts on inputs?

I fail to see how Kagi could even get approved for opting out of abuse monitoring for Gemini, let alone OpenAI.