Support configuring multiple model providers at a time

[justyns]

Right now we can only configure a single model in the SETTINGS page. We can change that to be an array of models instead, something like this:

ai:
  models:
  - name: ollama
    defaultTextModel: phi-2
    apiType: openai
    baseUrl: http://localhost:11434/v1
    requireAuth: false
  - name: openai
    apiType: openai
    defaultTextModel: gpt-4-0125-preview

models probably isn't the right name since each api could support multiple models (e.g. gpt-3 vs gpt-4). But part of this is so that I can support template prompts with a frontmatter like this:

aiprompt:
  description: "Describe the space script functionality you want and generate it"
  system: "You are an expert javascript developer.  Help the user develop new functionality for their personal note taking tool."
  model: gpt-4

Each templated prompt would be override the model it uses. Something like creating a new space script in my example above could go to openai/gpt4, but something like summarizing a personal note would go to a locally-hosted llm instead for privacy reasons.

[zefhemel]

I was just thinking about this too, would be great to have! You can use the editor.filterBox UI to allow people to easily switch

[justyns]

@zefhemel how do you feel about a plug updating SETTINGS automatically? I'm thinking of having a command that opens a filterBox with configured llms, then updating ai.selectedModel in SETTINGS to whatever the user selects.

[zefhemel]

That would be nice, but "restringifying" the YAML settings in place is likely restructure the settings and remove comments which users may not like.

[justyns]

I ended up going a different route and using clientStore.set and clientStore.get to set ai.selectedTextModel when it is changed. AFAICT this means it will be a client-specific setting like dark mode, and I don't have to worry about figuring out how to auto-update SETTINGS or a new config file