On start, we do a pi health check. This is half of what we need -- we also need to confirm they have a valid model ready to use.
For this, we want to have some kind of "model check" as well. In an app-start check, or maybe onboarding, we need to have them at least setup one model, and we have to verify it in some way.
The big issue right now is many people might be on Claude, and since we use pi, they wont be able to use their regular subscription. As such, for now, we will not focus on Claude users. Maybe note this in the README or somewhere, when we add a section to document our model selection onboarding. That basically, to use Claude with pi, they'd need to buy tokens.
And, as a result, we recommend other providers. We can have a stack rank or something, OpenAI, then maybe one or two others that allow the usual subscription based payment for usage with third-party harness like pi.
Do some deep thinking about this, consider edge cases, come up with a series of concerns and questions around this we need to explore before we build a basic onboarding flow to verify and ensure there is a working LLM model for use.
On start, we do a pi health check. This is half of what we need -- we also need to confirm they have a valid model ready to use.
For this, we want to have some kind of "model check" as well. In an app-start check, or maybe onboarding, we need to have them at least setup one model, and we have to verify it in some way.
The big issue right now is many people might be on Claude, and since we use pi, they wont be able to use their regular subscription. As such, for now, we will not focus on Claude users. Maybe note this in the README or somewhere, when we add a section to document our model selection onboarding. That basically, to use Claude with pi, they'd need to buy tokens.
And, as a result, we recommend other providers. We can have a stack rank or something, OpenAI, then maybe one or two others that allow the usual subscription based payment for usage with third-party harness like pi.
Do some deep thinking about this, consider edge cases, come up with a series of concerns and questions around this we need to explore before we build a basic onboarding flow to verify and ensure there is a working LLM model for use.