I've been thinking about all the use cases I've seen online for OpenAI and it got me thinking about payment models. Three I imagine will be popular are:

- Pay as you go: Website charges you per interaction, with the some/all of that cost going towards their OpenAI API usage

- Fixed cost, unlimited usage: Websites charge you once to use the service, and choose a price that reflects the average cost of a user's OpenAI api usage

- Freemium: Like the above, but the paid users also pay for the limited use that free users use.

However, all of these models require the user to submit their billing details to any website they want to interact with, which many users are understandably wary of and adds a lot of friction to trying new services (unless they follow the freemium model, which can be hard to implement and avoid misuse).

So, I'm proposing another option, very similar to how Metamask works (NB: Other than this analogy, my proposal has nothing to do with cryptocurrency - please lower your pitchforks):

- You insert your OpenAI API credentials to the Chrome extension - Any website can request to connect to the extension - When that request happens, the extension opens a window asking you how many credits or how much money you're willing to let the app use. - Once you've approved that, the app can make OpenAI calls through the extension as a proxy, and the extension will disallow requests after the allowed quota is reached.

This would have the advantages of:

- Letting the user safely test out AI-powered applications without worrying about the cost - Making it easier to create web apps that integrate with ChatGPT without having to pay a large upfront cost (let users pay for what they use and nothing more)

Other possible features could be:

- Grant a daily quota for apps that you keep coming back to - No-trust mode: See and approve every individual OpenAI request before it is sent - Audit log: see all requests ever made, with cost and attribution

Disadvantages and attack vectors:

- Websites could issue requests that break OpenAI rules and this transfers responsibility/consequences onto the end-user rather than the operator - It's another service users would need to trust. It could be open-source, but that doesn't preclude the possibility that a malicious version would be uploaded to the Chrome Web Store

I would love any feedback!