> For the complete documentation index, see [llms.txt](/llms.txt).
> A full single-fetch corpus is available at [llms-full.txt](/llms-full.txt).
---
title: "Datasets"
description: "Create eval datasets, publish immutable versions, import examples from runs or CSV, and deduplicate items."
last_verified: 2026-07-27
---


**Endpoints**: `GET/POST /api/v1/projects/:projectId/datasets`, versions, import, dedup, and examples sub-resources — see the endpoint list below
**Auth**: `X-API-KEY` or `Authorization: Bearer <token>` (required)
**Error envelope**: 4xx/5xx return `{"error": string, "message": string, "code": int, "hint"?: string}`


An eval **dataset** is a versioned collection of input/expected-output pairs used to run [experiments](/docs/api-reference/experiments.md). A dataset has a mutable draft and a history of immutable, published versions.

## Next steps

* [Experiments](/docs/api-reference/experiments.md): run a component or prompt against a published dataset version
* [Scorers](/docs/api-reference/scorers.md): define what "correct" means when an experiment runs against this dataset
* [Quality Cases](/docs/api-reference/quality-cases.md): turn a failing dataset item into a tracked regression