What problem does it solve?
This Skill solves the problem of scattered AI provider calls by defining a single routing layer where every model request goes through one controlled interface. It helps teams avoid client-side key exposure, inconsistent retry behavior, and ungoverned model usage.
Core Features & Use Cases
- Single routing choke point: Consolidates all model calls behind one internal gateway so credentials, model selection, cost limits, and failure handling are enforced consistently.
- Server-side credential custody: Ensures provider keys stay off the client bundle and are managed through secure server-side handling with rotation support.
- Deterministic routing and resilience: Selects models by task, cost, latency, and availability while supporting bounded retries, provider fallback, degraded responses, circuit breakers, and a kill switch.
- Telemetry at the call level: Defines per-request metrics such as model, token counts, estimated cost, latency, correlation id, and fallback usage for observability and chargeback.
- Use case: Use this Skill when replacing direct OpenAI and Anthropic calls with one governed AI gateway that enforces budgets and survives provider outages.
Quick Start
Ask the AI to design a server-side model routing layer that centralizes provider calls, keeps credentials out of the client, enforces budgets at one choke point, and includes fallback, telemetry, and a kill switch.
