Summary of Changes

Hello @nekomeowww, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request significantly enhances the application's interaction with Large Language Models by implementing a dynamic tool compatibility discovery system. This ensures that the application only attempts to use tools with models that explicitly support them, preventing potential errors and improving the overall stability and user experience. The changes span core LLM logic, UI integration, and testing infrastructure.

Highlights

• Dynamic Tool Compatibility Discovery: Introduced a robust mechanism within the llm store to dynamically discover whether a selected Large Language Model (LLM) supports tool usage. This is achieved by attempting API calls with and without tools enabled and analyzing specific error messages from providers like Ollama and OpenRouter. The results are cached to optimize performance.
• Conditional Tool Passing: The core stream function in the llm store has been updated to conditionally include tools in the request payload only if the active model has been discovered to support them. This prevents errors and improves robustness when interacting with models that do not have tool-calling capabilities.
• UI Integration for Tool Discovery: The InteractiveArea.vue components across stage-tamagotchi, stage-web, and stage-web/mobile applications now actively trigger the tool compatibility discovery whenever the user switches the active AI provider or model. This ensures that the system is always aware of the current model's capabilities.
• Testing and Dependency Updates: Added Vitest as a testing framework and introduced a new test file (llm.test.ts) to validate the tool compatibility discovery logic. Various @xsai/shared dependencies have been updated to newer versions, likely to support new error handling and other features.

Code Review

This pull request introduces a crucial feature for discovering tool compatibility with various providers, which significantly enhances the robustness of the application. The implementation is well-thought-out, especially with the addition of tests and caching for the discovery results. My review focuses on a few areas to improve maintainability and resilience. I've suggested refactoring some duplicated logic in the Vue components into a reusable composable. I also pointed out that the error handling in the core discovery logic is a bit brittle due to its reliance on string matching, and I've proposed a more robust approach. Finally, I've made a suggestion to simplify or at least clarify a complex custom promise utility function. Overall, these are great changes that move the project forward. Addressing these points will make the new functionality even more solid.

COOL