Why no keys

The hub never holds upstream LLM API keys.

• Sellers keep keys on their inference stack.
• Buyers pay on-chain, then call the seller endpoint with the grant.
• Compromise of the hub does not dump a pool of provider credentials.
• The rail stays a matcher + x402 coordinator, not a proxy LLM SaaS.

If you need the hub to "just call the model for you," that is a different product. CURB deliberately does not.