Importing Data
Import datasets from files and external sources into the Data Vault.
Bring your own reference data into the Data Vault — gold answers, expected tool calls, task seeds, or any tabular data you want to evaluate agent runs against or compare alongside collected results. Imported datasets sit next to the datasets collected from agent test runs.
Import from file
Upload a CSV, JSON, or JSONL file to create a new dataset.
- Navigate to Data Vault and click Import Dataset.
- Select the File Upload tab.
- Drag and drop your file or click to browse.
- Review the preview — column types are auto-detected and can be adjusted using the dropdown in each column header.
- Enter a dataset name.
- Click Import.
Supported formats
| Format | Details |
|---|---|
| CSV | Comma-separated values. First row is treated as headers. |
| JSON | Array of objects ([{...}, {...}]). |
| JSONL | JSON Lines — one JSON object per line. |
Limits
- Maximum file size: 50 MB
- Maximum rows: 100,000
If the file's columns match a known platform export (e.g., LangSmith, Braintrust, Scale AI, Labelbox), a badge is shown in the preview and the source platform is recorded on the dataset.
Import from HuggingFace
Import datasets directly from the HuggingFace Hub. Open Import Dataset, select the HuggingFace tab, search for a dataset, choose a split and configuration, and import it into the Data Vault.