A cache hit is a token bill of zero

Your support bot answers the same forty questions all day, and every answer bills tokens.

Point the bot's base URL at an exec script instead of the AI gateway: one config change. The script hashes each request; on a miss it forwards to the gateway, returns the completion, and stores it. On a hit it returns the stored completion, and the gateway never hears about it. No call, no tokens, nothing drawn from your AI Balance.

The bot's code didn't change. The cache policy: what to hash, how long to keep it: is a script you edit, not a feature you wait for a vendor to ship.

For repeated questions, the saving isn't a discount on tokens. A cache hit is a token bill of zero.