The real problem is that Bolt Local use a predefined list of models per provider, so that list should be updated every time the provider changes the status of the models (like for Perplexity), and Bolt has no way to validate or suggest them.

Right now, Bolt Local doesn't allows users to input any model name when setting up the Perplexity provider. It shows invalid models (e.g., llama-3.1-sonar-small-128k-online), it causes an error during API calls. Because these are no longer supported by the api calls by perplexity.

To improve user experience and prevent runtime errors, we should add:

• A dropdown menu or autocomplete input for supported model names
• Optional validation if a custom model is entered

This would help users avoid common mistakes and align with the Perplexity API's supported models.