A new way to bill AI agents just dropped: the x402 upto scheme ๐ค๐ธ...

How many tokens will an LLM generate? How much compute will an agent need?
With upto, the client authorizes a maximum amount, and the server settles for the actual amount used at the end of the request.
- LLM Inference: Agent authorizes up to $5, gets charged exactly $0.42 based on tokens streamed.
- Bandwidth: Pay precisely per byte transferred.
- Compute: Charge based on actual execution time.
In previous implementations, variable onchain billing requires locking funds in escrow or streaming contracts. Capital gets stuck.
With upto, agent funds remain untouched in their wallet, free to be used elsewhere until the exact moment of settlement.
You can either start with the official docs or CDP's facilitator docs:

