Datasets
datasets are awesome
The requests that broke last month are the best test set you will ever have. Cloudmind turns any trace into a dataset row in one click, so the hard cases stop coming back.
Built from what already happened
One click from a trace
Found a bad answer? Add it to a dataset with the expected output while you still remember what it should have said.
Filters become datasets
Any search over your traces can be saved as a living dataset that picks up new matching traffic.
Shared with your graders
Datasets are what evaluations run against, so the loop from "a user complained" to "this can never regress" is short.
Exportable
JSONL out, at any time, with no lock-in. It is your data.