Fix: LiteLLM API key field not showing in UI configuration

[jalateras]

Summary

• Fixed issue where LiteLLM API key input field was not appearing in the UI provider configuration modal
• Changed LITELLM_API_KEY's required flag from false to true to match other providers

Problem

As reported in #3953, users were unable to set an API key for LiteLLM because the input field did not show up in the Settings > Models > Configure Providers > LiteLLM dialog.

The root cause was that the UI only displays configuration fields that are marked as required: true, but LITELLM_API_KEY was incorrectly marked as required: false.

Solution

Updated the LiteLLM provider metadata to mark LITELLM_API_KEY as required, consistent with how all other providers (OpenAI, Anthropic, Google, etc.) handle their API keys.

Test plan

• Added unit test to verify LITELLM_API_KEY is marked as required
• Test passes: cargo test test_litellm_metadata_api_key_required
• Code formatted with cargo fmt
• Manual testing: Verify API key field now appears in UI

Fixes #3953

[jamadeo]

Thanks @jalateras , a couple of thoughts here:

• I'm surprised we only render the input if the key is required. @zanesq do you know why there isn't a way to have an optional input in the UI?
• Is it always required to have a LITELLM_API_KEY? I haven't used litellm but from a quick glance they seem to support running your proxy locally -- would this be a case where you wouldn't need to set it? what happens then?

[zanesq]

@jamadeo I think it was just a UX consideration at the time to limit the fields to only whats needed and keep it simple for the user to configure if nothing is needed. That being said I think it makes sense to put an optional field for everything and pass it along in case since params can change later also.

[jalateras]

Thanks for the review @jamadeo and @zanesq!

To address your questions:

1. Why UI only renders required fields: As @zanesq mentioned, this was a UX decision to keep configuration simple. However, I agree that showing optional fields makes sense for flexibility.

2. Is LITELLM_API_KEY always required?: Good catch! LiteLLM can indeed run locally without an API key (e.g., when using local Ollama models or as a proxy to other local services). The code already handles this - see litellm.rs lines 49-53 where it uses NoAuth when no API key is provided.

Recommendation

I suggest we proceed with this PR as an immediate fix because:

• It unblocks users who need to configure LiteLLM right now
• It maintains consistency with other providers (all mark API keys as required)
• The backend already gracefully handles missing API keys with NoAuth
• Users with local setups can enter a dummy value if needed

For the longer-term solution, I propose creating a follow-up issue to enhance the UI to show optional configuration fields. This could be implemented by:

1. Removing the filter that hides non-required fields
2. Adding visual indicators (asterisks for required, "(optional)" label for optional)
3. Updating validation to only enforce required fields

This way we deliver immediate value while planning for a more flexible configuration UI. What do you think?

[jalateras]

✅ E2E Testing Completed

I've successfully tested this fix end-to-end with LiteLLM:

Test Setup:

1. Started LiteLLM proxy:

# config.yaml
model_list:
  - model_name: openai-gpt-4o-mini
    litellm_params:
      model: openai/gpt-4o-mini
      api_key: sk-xxx

litellm_settings:
  drop_params: true

and to run

  docker run -p 4000:4000 \
    -v $(pwd)/litellm_config.yaml:/app/config.yaml \
    ghcr.io/berriai/litellm:main-latest \
    --config /app/config.yaml \
    --port 4000

3. Launched Goose UI using the fix from this PR
4. Configured LiteLLM provider in Settings → Models → Configure Providers → LiteLLM

Test Results:

• ✅ LITELLM_API_KEY field now appears in the configuration dialog (was missing before)
• ✅ Successfully entered and saved the API key through the UI
• ✅ Connected to LiteLLM proxy at http://localhost:4000
• ✅ Successfully sent messages through the chain: Goose UI → goosed → LiteLLM → OpenAI (gpt-4o-mini)
• ✅ Received responses correctly

Verification:

The fix works as intended - changing required: false to required: true makes the LITELLM_API_KEY field visible in the UI, allowing users to configure LiteLLM through the interface rather than having to manually edit config files.

Tested with:

• LiteLLM proxy serving OpenAI's gpt-4o-mini model
• macOS environment
• Current branch: fix/litellm-api-key-required

[jalateras]

Follow-up Work Needed

A separate PR should address the wider architectural issue of supporting truly optional configuration fields in the UI. This would involve:

• Modifying DefaultProviderSetupForm.tsx to show optional fields with proper visual indicators
• Adding "(optional)" labels and avoiding asterisks for non-required fields
• Updating validation logic to only enforce required fields
• Then reverting this change to make LITELLM_API_KEY optional again