Prompt caching cuts repeat tokens by up to 90%. Fallback keeps you up. Healing fixes broken JSON.   OpenRouter OpenRouter ( https://openrouter.ai/?utm_campaign=Onboarding+Flow+-+v4+April+2026&utm_content=A5+Cost+Features&utm_medium=email_action&utm_source=customer.io )   Hi botnews, Three OpenRouter features quietly do work for you in production. They are off by default on most platforms and on or one-flag-away here. Each one pays for itself within the first week on any workload that hits real traffic. View the docs ( https://openrouter.ai/docs?utm_source=email&utm_medium=lifecycle&utm_campaign=onboarding_v2&utm_content=a4_power_features&utm_campaign=Onboarding+Flow+-+v4+April+2026&utm_content=A5+Cost+Features&utm_medium=email_action&utm_source=customer.io )   PROMPT CACHING If you send the same system prompt or context window repeatedly, OpenRouter caches the prefix and bills cached tokens at up to 90 percent off. It runs automatically on supported models with no setup. The sa