01
Kylace
A model decision, backed by workload evidence.
Which model to run and when to change it
| Spec | Value |
|---|---|
| status | Pending |
| unit of work | The harness |
| cost is measured | Per accepted task |
| domain | kylace.com |
Overview
Kylace measures how AI models perform on your workload, recommends what to run, and updates the recommendation automatically.
The cheapest model is not necessarily the cheapest system, and the strongest model does not necessarily justify its higher price. What matters is what an accepted result costs.
The measurement uses your prompts, tools and real cases. Each candidate model, provider and setting is scored on speed, cost per accepted task and error rate. The recommendation says where a change is faster, cheaper or fails less often.