Ask HN: User-scoped AI tokens for safe client-side AI logic?

Consider a service providing user-scoped AI tokens. The service would allow you to configure usage limits for each user's tokens, allowing you to safely move AI logic to the client-side, simplifying your code.

As an example flow, you would register a user's unique secret with the service. Then, the service's client-side library would redirect all of the OpenAI requests to itself, and would also pass the unique user secret. If the corresponding user has no funding for the AI request, it would deny it.

Any interest?