pnd280 / complexity

⚡ Supercharge your Perplexity.ai
https://cplx.vercel.app
396 stars 25 forks source link

[Feature Request]: Dynamic Model Workflow / CPLX Dynamic Profiles #26

Open para-droid-ai opened 3 weeks ago

para-droid-ai commented 3 weeks ago

Feature Description

From: sickn233 — Today at 5:12 PM Hi, I don't know if my request is too ambitious, but would it be possible to implement an option that can automatically select the most suitable model depending on the type of request?

Use Case

Curious if we could have a "dynamic model selector" option or step in the workflow. But as things go, how does CPLX standarize on model use? seem like that would be most of the work for that, having a system dynamically routing requests to specific models.

I almost picture an entire CPLX Profile in place, for coding/long-form tasks/writing/etc. In this case, you'd setup your Profile with "preferences" for model use. That way it's less abstract, and still entirely user focused.

We would likely include default "profiles" but the idea is users can set it up to their needs, and then not worry about what model to use going forward, as it's being dynamically handled. You'd also be able to swap your CPLX Profile for different domain expert uses. This is heavily inspired from my testing of Highlight and how they manage dynamic output via user context.

Alternatives Considered

(used sickn233's OG comment, and my "thought-process" as Use Case. hoping this formatting works for the intent)

para-droid-ai commented 3 weeks ago

Per Highlight Rewrite:

[Feature Request]: Dynamic Model Workflow / CPLX Dynamic Profiles

Feature Description: Proposing an automated model selection system that can intelligently choose the most appropriate model based on request type.

Use Case: Implementation of a "dynamic model selector" workflow step to standardize model usage patterns. This would function like a CPLX Profile system, where users can configure profiles with preferences for different use cases. The system would include default profiles while allowing customization, automatically determining the optimal model for each request based on the configured profile settings.

The key benefits are:

tytrate commented 1 week ago

What do you think about this. it would not automatically determine which model is best. You could send to all of the models and then write prompts to consolidate the output with redundancy removed and contradictions called out.

https://github.com/pnd280/complexity/issues/40