Insights from Testril
We write about onchain data, AI agents, and how machines buy data from each other.
Why AI Agents Waste 97% of Their Tokens on Raw Data
We measured what 18 stats cost an agent computing them from raw rows instead of asking for the finished answer. 12,476 tokens versus 395, a 97% cut.
ThesisWhy AI agents don't need raw blockchain data
Feeding an agent raw blocks, logs, and traces is slow and expensive. What an agent needs is the answer, and a way to check that it is right.
EngineeringProvenance by default: making onchain answers verifiable
Every value Testril returns traces back to the blocks it came from and the code that produced it. That is what lets an agent trust an answer.
ProductSee the price before anything runs
Testril quotes every request before it runs, so an agent sees the cost first and nothing happens until it agrees to pay.
PaymentsWhat x402 and MPP mean for machine payments
When an agent pays for each query, the payment rail matters. A short explanation of x402, MPP, and per-request pricing.