Unhardcoded drops open-source model router

Unhardcoded targets the fragile practice of hardcoding model names like 'gpt-4o' directly into application code, providing a much-needed abstraction layer for production LLM deployments. By putting policy decisions outside application logic, it makes AI apps model-agnostic and far easier to maintain as frontier models update.

• –Decouples application logic from specific LLM providers, allowing seamless upgrades and cost optimizations without redeploying code.
• –Open-source approach using developers' own API keys avoids vendor lock-in and middleman markups.
• –Supports fallback chains and capability-based routing, such as routing only to vision or JSON-compatible models.
• –Produces deterministic, replayable decision traces that simplify debugging and auditing of model selection logic.