Dataset Runs
Connect a dataset to Playground variables and run batch execution across every row to systematically test your configuration.
Dataset Runs
Instead of testing one input at a time, you can connect a dataset to the Playground and run your prompt configuration against every row in bulk. This is useful for systematic regression testing before promoting a prompt change.
Running a Dataset Through a Configuration
Add a variable to your prompt
In your prompt messages, add a Handlebars variable such as {{input}}. The Variables panel on the left shows a row for each detected variable.
Switch the variable type to Dataset
In the Variables panel, change the type dropdown for the variable from Text to Dataset.
Select a dataset
A dataset picker appears. Choose the dataset you want to use from the list of datasets in your project.
Select a version
Pick the dataset version to run against: Latest, Production, or a specific numbered version.
Map a column to the variable
Select which dataset column should supply the value for {{input}} (e.g. map the question column to {{input}}). Repeat for each variable if you have multiple.
Set the row limit
Enter the number of rows to run (defaults to the full dataset size). Reduce this for quick spot checks.
Run the dataset
Click Run (or Run All if you have multiple windows). The Playground calls the model once per row, substituting that row's column value into the prompt. A results table below the windows shows each row's output as it completes.
Review results row by row
Click any row in the results table to open a detail view showing the exact prompt sent and the model's response for that row.
When multiple windows are open, the results table shows one column per window so you can compare outputs from different models or prompt variants across the same dataset rows.
Logging Results as an Experiment
After a dataset run completes you can save the results as a formal experiment for long-term tracking. Click Log as Experiment in the results area, give each window's run a name, and confirm. The runs will appear in the Experiments section of your project.