[custom endpoint] Add support for getting auth token from command line

[sa-]

At our company we have an OpenAI API compatible model router deployed which can only be accessed via short lived tokens.

This means we can't put a long-lived API key in the configuration since long-lived tokens cannot be generated.

We would be able to use this if we provide a custom bash command to get an auth token like get-auth-token --name vs-code-ai

[swatDong]

It's able to use environment variable when adding a model to "My Model" list.

E.g., for Open AI models,

• Create environment variable OPENAI_API_KEY on your machine

• (Close and reopen VSCode instances to ensure the env var works in VSC)

• When adding Open AI model, you'll see an option "Use $OPENAI_API_KEY"
  

• After adding model like this, AI Toolkit will call the model by referencing the value from your env var.

BTW, Antropic model uses ANTHROPIC_API_KEY, Google model uses GEMINI_API_KEY

(Let me also add these steps to the doc)

[sa-]

That's good to know! Although it doesn't quite solve the issue as we would have to quit and reopen VS code every few hours from the command line like

OPENAI_API_KEY=$(get-auth-token) code

It's just inconvenient enough that people would be discouraged from using this

[drewmacphee]

We too have moved to oauth only for aoai access, specifically Az cli Entra ID. It would be nice if that were supported.
Thanks.

[swatDong]

Thanks all for the detailed scenarios. Added to our backlog.
FYI @sffamily

[sffamily]

@sa- Your use case is a valid scenario that we've added to our feature backlog and will investigate and plan it for our next release.