Model configuration (env variables)

Background

Today we only allow a user-friendly experience to configure model provider and model through the CLI (#352). However there are several gaps

we don't have a similar experience in the Web UI, so users must pass through the CLI or use the settings tab
we only enable a single configuration globally at a time (1 LLM provider + 1 api key + 1 model)
if a users changes the model configuration, it overrides the previous one, so there is no toggling.

Can we enable users to have a similarly friendly first-time experience of setting up an LLM provider and selecting a model in the GUI that we have in the CLI?
The CLI currently is a "wizard" type experience. What if a user just wants to enter the three values?
Can we store multiple API keys so they can switch providers and models more easily?
Where does a user inspect their stored variables/configs? (CLI and GUI)
If we had multiple configurations stored, where would a user set default?
Can we store multiple API keys so they can switch models across providers?
Do we include a required (or optional) definition of recommended model for agents and reflect that in the UI (recommended tag or similar)
Can a user test different models (i.e. swap among saved configurations) at model runtime?
Do we allow manually changing the 3 environment variables related to the provider and model setup (as we currently have in the settings GUI)?
How do we create the equivalent experience CLI / GUI with transparency, simplifying the setup for single or multiple LLM providers (if possible)

I think we need to blow this one out and turn it into a discussion vs. a feature. Lots to unpack here.