> ## Documentation Index
> Fetch the complete documentation index at: https://docs.keywordsai.co/llms.txt
> Use this file to discover all available pages before exploring further.

# Create datasets in the UI

> A guided walkthrough to create datasets from existing logs (sampling) or create an empty dataset and add logs via JSON/CSV.

## What is a dataset?

A **dataset** is a curated collection of logs (inputs/outputs + metadata) that you can evaluate, annotate, and use to power Experiments.

## Resources

* [Create datasets via API](/documentation/products/dataset/via_api)
* [Datasets API reference](/api-endpoints/evaluate/datasets/list)
* [CSV template for importing dataset logs](https://drive.google.com/file/d/10U_nYZxCd_uOYyuZob9x0gE6McJPdYb8/view?usp=drive_link)

## Steps to use

<Tabs>
  <Tab title="Sampling existing logs">
    <Steps>
      <Step title="Step 1: Go to Datasets and click Create">
        Open the Keywords AI platform, navigate to **Datasets**, then click **Create dataset**.

        <Frame className="rounded-md">
          <img src="https://keywordsai-static.s3.us-east-1.amazonaws.com/docs/documentation/products/dataset/create.png" alt="Datasets page with Create dataset button" />
        </Frame>
      </Step>

      <Step title="Step 2: Choose “From logs (sampling)”">
        In the creation modal, choose dataset type **From logs (sampling)**.

        <Frame className="rounded-md">
          <img src="https://keywordsai-static.s3.us-east-1.amazonaws.com/docs/documentation/products/dataset/fromlogs_sampling.png" alt="Create dataset modal with dataset type set to From logs (sampling)" />
        </Frame>
      </Step>

      <Step title="Step 3: Set filters, time range, and sampling percentage">
        Configure:

        * **Filters**: narrow down which logs to include (e.g. `status_code = 200`, a specific `model`, metadata filters)
        * **Time range**: the window of logs to sample from
        * **Sampling percentage**: how much of matching logs to include

        <Frame className="rounded-md">
          <img src="https://keywordsai-static.s3.us-east-1.amazonaws.com/docs/documentation/products/dataset/filters.png" alt="Dataset from logs configuration showing filters, time range, and sampling percentage" />
        </Frame>
      </Step>

      <Step title="Step 4: Create, wait for ingestion, and verify logs">
        Click **Create**. The dataset will begin processing while logs are being added.

        Depending on the size of the time range and filters, it may take some time until the dataset is fully populated.

        When it finishes, open the dataset and confirm:

        * `log_count` is non-zero
        * logs appear in the dataset logs table

        <Frame className="rounded-md">
          <img src="https://keywordsai-static.s3.us-east-1.amazonaws.com/docs/documentation/products/dataset/see_logs.png" alt="Dataset detail page showing dataset logs table populated" />
        </Frame>
      </Step>
    </Steps>
  </Tab>

  <Tab title="Create empty + add logs (JSON/CSV)">
    <Steps>
      <Step title="Step 1: Create an empty dataset">
        Open the Keywords AI platform, navigate to **Datasets**, then click **Create dataset**.

        Choose **Create empty**. This creates the dataset immediately (no background sampling job).

        <Frame className="rounded-md">
          <img src="https://keywordsai-static.s3.us-east-1.amazonaws.com/docs/documentation/products/dataset/create.png" alt="Create dataset modal with Create empty option selected" />
        </Frame>
      </Step>

      <Step title="Step 2: Add logs by inserting JSON (Input + Output)">
        Inside the dataset, click **Insert new log** and provide a JSON log with **input** and **output**.

        <Frame className="rounded-md">
          <img src="https://keywordsai-static.s3.us-east-1.amazonaws.com/docs/documentation/products/dataset/insert_json.png" alt="Insert new log UI with JSON editor for input and output" />
        </Frame>

        <Tip>
          Datasets store end-to-end workflow I/O, so `input` and `output` can be any JSON structure.
        </Tip>
      </Step>

      <Step title="Step 3 (alternative): Import logs from CSV">
        If you already have a batch of examples, click **Import from CSV** and upload a CSV file.

        Use this template to get started:

        * [Google Drive CSV template](https://drive.google.com/file/d/10U_nYZxCd_uOYyuZob9x0gE6McJPdYb8/view?usp=drive_link)

        <Frame className="rounded-md">
          <img src="https://keywordsai-static.s3.us-east-1.amazonaws.com/docs/documentation/products/dataset/csv.png" alt="Import from CSV modal for dataset logs" />
        </Frame>
      </Step>
    </Steps>
  </Tab>
</Tabs>