> For the complete documentation index, see [llms.txt](https://karini-ai.gitbook.io/karini-ai-documentation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://karini-ai.gitbook.io/karini-ai-documentation/dataset-export.md).

# Dataset Export

### **Overview**

**Dataset Exports** enables users to generate structured datasets from **Copilot** conversations and **Webhook Recipe** execution history. These datasets can be used for analytics, evaluation, auditing, optimization, and model fine-tuning workflows. Exports are stored securely within the platform and can be previewed in the application or downloaded for offline use.

Dataset Exports are accessible from both:

* [**Copilot History**](/karini-ai-documentation/copilots.md#copilot-history)
* [**Recipe History**](/karini-ai-documentation/recipes/workflow-recipe/webhook-recipe.md#webhook-history-ui)

This feature provides a comprehensive set of capabilities that enable users to generate, manage, and consume structured datasets .

### **Exported Dataset Contents**

Dataset Exports retrieves run-level historical data from available Copilot and Recipe history records and outputs it in a structured format suitable for downstream consumption.

**Included Data** 

* **Prompts / Questions:** User queries or prompt text captured during execution
* **Inputs:** System inputs and user-provided inputs used for the run
* **Responses / Outputs:** Generated responses produced by the copilot or recipe execution
* **Execution Status:** Success/failure state and completion indicators
* **File References:** Links/identifiers for attached files associated with the run
* **Metadata:** Supporting attributes such as timestamps, run identifiers, and other trace fields

### Copilot History: Dataset Export

To create a **Dataset Export** from **Copilot History**, use the steps below to locate prior runs, select the export scope, and generate the dataset.

* **Export from Copilot History**
  * Navigate to **Copilot**
  * Select the target **Copilot**
  * Click the **Action** button for the selected Copilot
  * Select **View history**
    * This opens the **History (Activity Dashboard)** page, where each row represents a prior run and may include fields such as **Request ID**, **Question**, **Answer**, **Error**, and **Files**
  * **Select runs to export (Optional)**
    * Use the **row checkboxes** to export specific runs
    * If no rows are selected, control the export scope using **Limit** (exports the most recent *N* runs)
  * **Generate the export**
    * On the **History** page, click **Action**
    * Select **Export dataset**
  * **Configure export settings**
    * **Export Name:** Provide a descriptive dataset name
    * **Purpose:** Select the export intent. Typical options include:
      * Export Dataset
      * Recipe Evaluation
      * Node Evaluation
      * Node Optimization
      * Finetuning
    * **Nodes:** Required for node-based purposes (e.g., **Node Evaluation**). Select the node to evaluate/export
    * **Limit:** Set the maximum number of runs to include
  * **Finalize**
    * Click **Proceed** to generate the export

<figure><img src="/files/1gN1t6zwOyGFCue4I8Cv" alt=""><figcaption></figcaption></figure>

### Webhook History: Dataset Export

* To create a **Dataset Export** from **Webhook History**, use the steps below to locate prior runs, select the export scope, and generate the dataset.
  * **Export from Webhook History**
    * Navigate to **Recipe**
    * Select the target **Webhook recipe**
    * Click the **Action** button for the selected recipe
    * Select **Webhook history**
      * This opens the **History (Activity Dashboard)** page, where each row represents a prior run and may include fields such as **Request ID**, **Question**, **Answer**, **Error**, and **Files**
  * **Select runs to export (Optional)**
    * Use the **row checkboxes** to export specific runs
    * If no rows are selected, control the export scope using **Limit** (exports the most recent *N* runs)
  * **Generate the export**
    * On the **History** page, click **Action**
    * Select **Export dataset**
  * **Configure export settings**
    * **Export Name:** Provide a descriptive dataset name
    * **Purpose:** Select the export intent. Typical options include:
      * Export Dataset
      * Recipe Evaluation
      * Node Evaluation
      * Node Optimization
      * Finetuning
    * **Nodes:** Required for node-based purposes (e.g., **Node Evaluation**). Select the node to evaluate/export
    * **Limit:** Set the maximum number of runs to include
  * **Finalize**
    * Click **Proceed** to generate the export

<figure><img src="/files/y2zCcE8vQ2njic49hIbt" alt=""><figcaption></figcaption></figure>

### Export Purpose and Output Formatting

When creating an export, the user must choose an **Export Purpose**. The selected purpose determines how records are filtered and how the/or data is structured for the intended workflow.

| Export Purpose    | Intended Use                                              | Output Characteristics                                                                      |
| ----------------- | --------------------------------------------------------- | ------------------------------------------------------------------------------------------- |
| Export Dataset    | General reporting, analysis, archiving                    | Generic historical interaction/execution dataset                                            |
| Recipe Evaluation | End-to-end evaluation of Recipe performance               | Evaluation-oriented structure to compare quality across runs/versions                       |
| Node Evaluation   | Validation of a specific node in a Recipe/Copilot flow    | Data scoped to a node to isolate behavior and performance                                   |
| Node Optimization | Iterative improvement of node prompts/logic/configuration | Data shaped for tuning decisions using real traces                                          |
| Fine-tuning       | Model training/adaptation workflows                       | Fine-tuning-friendly format (for example, JSONL) with structured message pairs and metadata |

### Export Management and Tracking

A dedicated **Dataset Exports** view is available for tracking and managing all generated exports. Each export entry displays essential metadata, such as export name, export purpose/type, node information (when applicable), record count, creator, and creation timestamp. From this view, users can preview an export using **View** or retrieve it using **Download**.

{% hint style="info" %}
**Note:** Recipe Evaluation exports are downloaded to the user’s local system.
{% endhint %}

<figure><img src="/files/svdwNkMs0o2wC6trZbNI" alt=""><figcaption></figcaption></figure>