main ← fix-skip-failure-tracking-for-unsaved-llm-models
opened 09:31PM - 16 Dec 25 UTC
When testing a new LLM configuration in the admin panel, if the test fails (e.g.…, invalid API key), users see "Target can't be blank" instead of the actual error from the LLM provider.
This happens because `track_failures` attempts to create a ProblemCheckTracker using `llm_model.id` as the target, but unsaved models have `id = nil`, which fails the `validates :target, presence: true` validation.
The fix adds a `new_record?` check to skip failure tracking for unsaved models, consistent with the existing checks for `blank?` and `seeded?`.
Ref - https://meta.discourse.org/t/391497