vividfog
commented
9 months ago The UX was fine. Some notes:
- Asking for OpenAI key when I came with the idea that I'll use a local model for this. Perhaps "press enter to not use OpenAI" or something would fix that.
- Installing the alias didn't work, it just silently failed to add the alias there. Perhaps the docs could say out loud to make alias "??"=clipea, because for target users for a tool like this, they would then know what's going to happen behind the scenes, or fix/do it quickly themselves.
- Being explicit about applying "llm models default" would be good, so there's no need to watch server logs to verify that it indeed does.
- Some sort of wizard would indeed be nice, especially if it takes a closer walkthrough of what model to use, what template to apply (if any) ...etc.
- Having the option to use a stronger model is good, so I like that "4" remains accessible even after using llm defaults to something else. Appreciated.
- Being a pedantic, perhaps the Clipea prompt could become customizable down the road. A clip would be enough for me.
- Ability to specify "--no-stream" as part of defaults is useful for some models/proxies that don't support streaming. This could also be expected from the underlying llm though.
- Generally, cross-referencing some parts of the llm library, in particular for customizing model choices and then applying them for clipea, could be a nice enhancement down the road.
- A set of known-good models to use with Clipea via LLM, perhaps related to previous point, that'd help a newcomer to make a decision, perhaps get excited too.
That's what stuck my mind upon unboxing. Hope this helps.
First use wizard. disclaimer user preferences setup