feat: configure dynamic providers via .env

[mrsimpson]

Motivation

Up to now, some providers couldn't be configured via the environment properly. The list of models remained empty

What this PR contains

Previously, the list of models was retrieved from the front-end. If the list wasn't hard coded (like for many providers), the system tried to retrieve the models via the /models route of the provider.
As the environment configuration which includes the API keys is intentionally not shared with the client, this request couldn't complete and failed silently on the client.
This affected openai like and openrouter once they required api keys.

This PR implements a more robust and flexible approach to retrieving model lists by moving the initialization logic to the backend, enabling dynamic model discovery and configuration across different providers.

Key Changes

1. Model List Retrieval
   Replaced static MODEL_LIST with dynamic backend-driven model fetching
   Created a new /api/models endpoint to serve model information
   Implemented useModels hook for frontend model access
2. Provider and Model Management
   Refactored model initialization to use LLMManager
   Added more flexible provider and model configuration
   Improved handling of API keys and provider settings
   Technical Details
   LLMManager

Key implementation details:

• Enhanced updateModelList method to handle dynamic model discovery
• Improved provider configuration detection
• Added method to retrieve providers and default provider

API Endpoints

/api/models now returns:

• Complete model list
• Available providers
• Default provider configuration

Related

#1035

[thecodacus]

I have added my comments, I didn't want to add another api endpoint but I guess thats the only way to have both apikeys from UI and from env since we should not expose apikeys set on env to UI.

the only change i think is needed is passing the UI data from cookies so that it works for every scenarios env/UI

[thecodacus]

also there is a section when apiKey changes in UI we update the models for that provider.
BaseChat.ts file

but since we are using endpoint to get the models list. shall we change this also?

[mrsimpson]

Thanks for the catch with the override, @thecodacus !

I re-enabled the provisioning of apiKeys from the UI.

In the video, you can see the api-key being loaded from the env, then overridden with an invalid api-key (the models-list becomes blank as earlier), then override the api key with the valid one from the UI.
https://github.com/user-attachments/assets/ed28b841-7f84-45fb-8602-163b5b786753

Hope this is fine now.

[thecodacus]

In the video, you can see the api-key being loaded from the env, then overridden with an invalid api-key (the models-list becomes blank as earlier), then override the api key with the valid one from the UI.
https://github.com/user-attachments/assets/ed28b841-7f84-45fb-8602-163b5b786753

but i believe in this case its not using the /models endpoint but using the browser to call the providers. can you confirm that by checking the browser console ?

[mrsimpson]

@thecodacus

I now transport all the apikeys and settings via cookies. It still makes me shiver to have an unencrypted textfile with credentials transported as plain text, but ... maybe I have been too corporate all my dev life ;)

you asked

but i believe in this case its not using the /models endpoint but using the browser to call the providers. can you confirm that by checking the browser console ?

I verified manually on the dev tools: When changing the apiKey on the ui, a new request to /models is performed.

[thecodacus]

all looks good to me. I will just run it once then can merge it

[mrsimpson]

@thecodacus please let me merge it if it looks ok for you. I want to make sure I pushed the latest changes, but I'm afk now

Edit: all good, it was just the mobile GitHub app not refreshing 🤦