Why We Built a Self-Healing AI Gateway: Architecting for Provider Instability

1. The Fragility of the "Wrapper" Era: Why openai.chat.completions is a single point of failure.
2. Native Infrastructure vs. Shims: Why we abandoned SDK shims for native Go implementations of Google and Groq protocols.
3. The Health-Check Loop: How Nexus uses a background goroutine to monitor provider latency and error rates.
4. Autonomous Re-routing: The logic behind switching from a primary model to a secondary "Speed" model (Groq) when latency spikes.
5. Conclusion: Why "Sovereign Infrastructure" is the only way to scale AI to the enterprise.