Integrate hosted and open-weight models using APIs and local runtimes, with streaming, retries, batching, rate limits, model routing, and fallback behavior.
