Datasets

Overview

Datasets are collections of structured test data used to evaluate and test your prompts across multiple scenarios. They contain rows of data that can include chat messages, variables, or both.

Dataset detail page showing tabular data editor

Creating a Dataset

Using Examples

Click New Dataset
Select Use Example
Enter a dataset name
Choose an example (Q&A, Chat, or Mixed Format)
Click Create Dataset

Uploading CSV

Click New Dataset
Select Upload CSV
Enter a dataset name
Select your CSV file
Choose the separator (comma, semicolon, tab, or pipe)
Click Create Dataset

Starting Empty

Click New Dataset
Select Start Empty
Enter a dataset name
Click Create Dataset
Add rows and columns manually in the editor

Data Format

Messages Format

[
  {"role": "user", "content": "What is the capital of France?"},
  {"role": "assistant", "content": "The capital of France is Paris."}
]

Variables Format

{
  "question": "geography",
  "expected_answer": "Paris",
  "difficulty": "easy"
}

Editing Datasets

Click directly in any cell to edit
Use Edit messages or Edit variables buttons for JSON data
Click + at the bottom to add rows
Click + in the header to add columns
Use Import CSV to add more data
Click Download CSV to export

Best Practices

Use descriptive names
Start with examples to understand the format
Organize by use case (edge cases, happy paths, error handling)
Include expected outputs for evaluation
Keep datasets focused and manageable
Export regularly for backups