Skip to content

Quality

Milimarty edited this page Oct 15, 2026 · 1 revision

Quality

Build › Quality shows how well keel's flows work with each model. keel runs small eval projects (in content/evals/) through the flows you pick, in run mode auto, one case at a time, and scores each run from 0 to 100: did it reach the end, how often was something sent back, and did it stay within the token estimate. Pick one or two models and press Run now, or turn on Every night at …. When a score falls by 15 points or more since the run before, the page shows it in red: check the last change to a prompt, an agent, a skill or a model. The runs' tokens count in the budget.


From docs/GUIDE.md · All pages

Clone this wiki locally