An AI hardware buying Agent that checks real specs and rules

On October 3, 2026, DEV Community published a write-up of a project submitted to the Sanity Challenge under Path One: Ship an Agent That Queries Real Content. The post's title describes it as an AI hardware-buying agent that queries real content, with a focus on compute paths and rule validation. The project is an agent built to answer one practical question: "what computer do I need to run this AI model?" According to the write-up, its answers are not limited to a spec sheet — it may conclude "don't buy, rent," or tell a user "you already have enough."

What sets the agent apart, per the description, is where its knowledge comes from. It does not read a hardware catalog. Instead, it queries a Sanity dataset made up of criteria: formulas covering weights memory, KV cache, speed ceiling, and the breakeven point between buying a machine and renting one, along with rules that can be checked by machine. Those criteria are what the agent reasons over when it responds. The write-up also states that the project lays out 11 solution paths.

That is the extent of what has been reported so far: a contest submission and the author's own description of how the agent works. No evaluation results, benchmarks, rankings, or further updates appear in the material. The entry remains at the stage of a published submission.