{"name":"ai-compute","description":"Hourly GPU rental term structure: smoothed spot, 1w, and 1m marks in USD per GPU-hour, collected from public pricing sources. when an agent should use this: it needs a GPU compute mark, term structure, or hourly history to compare with power prices or a later spot.","url":"https://ai-compute-production.up.railway.app","version":"1.0.0","provider":{"organization":"ai-compute"},"capabilities":{"streaming":false},"defaultInputModes":["text/plain"],"defaultOutputModes":["application/json"],"skills":[{"id":"spot","name":"/v1/compute/spot","description":"Smoothed spot GPU rental marks in USD per GPU-hour by canonical GPU type. when an agent should use this: it needs the current on-demand or interruptible spot mark and does not need the term curve or a history window.","tags":["gpu","compute","spot","rental","mark"]},{"id":"term","name":"/v1/compute/term","description":"Current GPU compute term structure: spot, 1w, and 1m, by canonical GPU type. when an agent should use this: it needs the curve, including trailing tenors labeled trailing_median, rather than a single spot print or a long history.","tags":["gpu","compute","term","curve","rental"]},{"id":"history","name":"/v1/compute/history","description":"Hourly GPU rental mark history for correlation against an RTO, spread, or later spot. when an agent should use this: it needs a past window of smoothed marks, not the latest print alone.","tags":["gpu","compute","history","hourly","mark"]}]}