cuyamaca-llm-abstraction

To manage multiple LLM providers, you can use a dual-slot architecture that separates code generation from runtime control, allowing you to configure distinct providers like Ollama or OpenAI for each specific inference task.

To securely store API keys when coordinating multiple LLM providers, you can utilize the OS keychain workflow, which safely manages credentials for services like Anthropic, Google, and Mistral without exposing them in your configuration files.

You cannot use a non-multimodal model for runtime control, because the dual-slot architecture enforces multimodal capability checks specifically for the runtime slot to ensure safe and versatile live operation.

The best way to verify LLM provider health and validate slot configurations is by running built-in discovery and health-check commands, which confirm that your code and runtime model paths are operating safely.

Running this multi-provider LLM abstraction requires no specific external dependencies, allowing you to directly integrate the model manager and provider traits into your Rust or Tauri environment for flexible deployment.

You need separate slots for code generation and runtime inference to isolate sketch generation from live control, enabling independent provider configuration, flexible experimentation, and optimized deployment for each specific task.