# Cortex Project Rules

You are working on the Cortex AI Hub. Adhere strictly to the following architectural rules:

## Absolutely NO Hardcoding Model Limits or API Keys
Under no circumstances are you allowed to hardcode fallback values for LLM token limits (e.g., `1048576` for Gemini or `128000` for DeepSeek) or API keys.
1. The orchestration engine must rely **solely** on `litellm`'s dynamic model info registry or database preferences.
2. If a valid model or limit cannot be dynamically resolved, the code MUST raise a `ValueError` with an honest, actionable exception pointing the admin to the UI instead of masking the setup error with a silent fallback.
3. This applies to all Python files, especially in `/app/ai-hub/app/core/providers/`.