[mingweiliao]

Summary

Replaces the hardcoded LlmModel enum in the sample agent with dynamic model discovery from configured LLM providers, so newly released models are available automatically without requiring code changes.

Key Changes

• Dynamic LlmModel representation:
  • Converted LlmModel from a hardcoded enum class to a @Serializable data class LlmModel(val providerName: LlmProviderName, val modelName: String).
• Provider model discovery (LlmProvider.getAvailableModels):
  • Added suspend fun getAvailableModels(apiKey: String?): List<LlmModel> to LlmProvider.
  • Implemented GeminiProviderImpl.getAvailableModels to query GET https://generativelanguage.googleapis.com/v1beta/models?pageSize=100 and filter for gemini-* models supporting generateContent.
• UI & ViewModel updates:
  • Updated AgentDemoViewModel to fetch available models across providers when API keys change and automatically create an initial thread once models are available.
  • Made AgentUiState.Loaded.currentThread nullable so the app loads cleanly instead of staying on the loading spinner when no API key or thread exists yet.
  • Updated ModelDropdown in AgentDemoScreen to group models by LlmProviderName and handle empty model/thread states gracefully.
• Screenshot & unit/integration tests:
  • Fixed screenshot test reference directory paths (com/example/appfunctionsstudio -> com/example/appfunctions/agent) and added/updated previews for loading and no-thread states.
  • Added unit and integration tests for getAvailableModels and updated existing test suites.

Test

• ./gradlew spotlessCheck test validateScreenshotTest
• ./gradlew connectedStandardDebugAndroidTest