Skip to main content
Use Playground when you want a fast UI loop: load a prompt, run it on data, score with evaluators, compare runs, and keep the best setup.

​
Run an experiment

1

Load a prompt and dataset

Open Playground, load your prompt, then select a dataset (or replay a production span).
2

Attach evaluators

Add evaluators so each output is scored automatically.
3

Run and inspect

Click Run, then open View Experiment to review outputs, latency, tokens, and evaluator scores.

​
Compare experiments

Once you have multiple runs on the same dataset, open Compare Experiments to inspect:
  • Output differences
  • Evaluator deltas
  • Summary metrics by run
  • Regressions vs baseline
If you need full programmatic workflows, use Experiment in code.

​
Playground views

A Playground View is a named snapshot of your Playground session. It stores:
  • Prompt messages and tools
  • Model and parameter configuration
  • Dataset or span context
  • Generated results and scores
Use views to save progress, share a setup, or branch experiments without rebuilding from scratch.
Playgrounds list with options to edit name, duplicate, or delete saved playground views