ChatGPT Enterprise data source

This topic provides details on how to capture OpenAI's ChatGPT Enterprise with Collect.

This documentation contains references to third-party software, or technologies. While efforts are made to keep third-party references updated, the images, documentation, or guidance in this topic may not accurately represent the current behavior or user interfaces of the third-party software. For more considerations regarding third-party software, such as copyright and ownership, see Terms of Use.

Considerations

Note the following considerations about this data source:

This connector requires a ChatGPT Enterprise subscription.
Collect retrieves ChatGPT data through OpenAI's Compliance API, which exposes data on a rolling 30-day basis. Collect retrieves data generated within the past 30 days.OpenAI's Compliance API only keeps data for 30 days. Collect can only retrieve ChatGPT data from the past 30 days.
- OpenAI retains deleted data for no more than 30 days, unless a legal obligation requires longer retention.
- Temporary chats, sometimes called ephemeral chats, do not appear in the user's history, do not update memory, and are not used for model training. OpenAI retains them for up to 30 days, and the Compliance API captures them during that window. Collect includes temporary chats.
- To build a complete history, schedule regular and ongoing collections so you capture data while it remains within the 30-day window.
Before your first collection, confirm that your organization enabled the Compliance API in its OpenAI account. This setting supports retrieval of both persistent and temporary conversations.
For date criteria, ChatGPT Enterprise supports a start date. Collect retrieves all prompts and responses from the date you specify through the current date.
If your organization has multiple ChatGPT Enterprise workspaces, you must configure a separate data source for each workspace in the Collect application. Name each data source to identify the workspace connects to.
Relativity automatically converts collected conversation prompts and responses to Relativity's short message format (RSMF). For more information, see The Relativity Short Message Format (RSMF).
ChatGPT metadata is collected. For more information, seeCollected metadata.

Prerequisites

Complete the following before setting up a ChatGPT data source in Relativity:

You must contact OpenAI to get an API Key before configuring your ChatGPT Enterprise data source in Collect.
- You can use the same API Key for configuring multiple ChatGPT Enterprise data source connections.
- For more information, please see OpenAI documentation on their Compliance APIs for Enterprise customers.
Confirm that the Compliance API is enabled in your organization's OpenAI account to support retrieval of both persistent and temporary conversations.

Once you complete the prerequisites, you can begin creating the data source.

Collection size

ChatGPT Enterprise collections scale with conversation volume and duration.

Category	Details
Hard limits (enforced)	No explicit hard cap published
Planning thresholds (guidance)	High-usage enterprise tenants
When to split	Long date ranges
Notes	Conversion-based scaling

Focus on limiting time ranges in high‑usage environments to control collection size.

Collected metadata

As of the current integration with OpenAI’s Compliance API, the metadata fields are captured and included in collections:

The following metadata fields are captured from ChatGPT Enterprise conversations via the Compliance API and included in Collect:

Metadata	Collected
Conersation ID	Always
Conversation title	Always
User ID	Always
User name	When user is logged in.
User email	When user is logged in.
Timestamp of prompt	Always
Timestamp of response	Always
Prompt text	Always
Response text	Always
Edit history	When a prompt was edited before re-execution.
Source attributions	URLs of any publicly cited websites when browsing was enabled.
Attachments	Uploaded files submitted by users during the conversation. For example, PDFs, DOCX, images, CSV files. Generated files. For example, documents or images created by ChatGPT in response to prompts.

These fields are extracted for both persistent and ephemeral (temporary) conversations, as long as they fall within OpenAI’s 30-day retention window and meet your search criteria in Collect.

Creating the data source

Use the following procedure to connect the ChatGPT Enterprise workspace data source to Collect.

In RelativityOne:

Navigate to Collection Admin within Collect Admin of Set Up
Click the New Collection Source Instance button.
- Name—Enter in a unique name for the data source.
- Type—Select the ChatGPT Enterprise data source.
- Settings—enter the required information in the Settings fields. For more information, see Settings fields.
Click Save. The data source displays on the Collection Admin page.

Settings fields

To connect Relativity to a ChatGPT Enterprise workspace, you need to gather and enter the information for the following fields.

API Key—enter the API Key that OpenAI provided to your organization. You can use the same API Key when configuring multiple ChatGPT Enterprise data sources from which you wish to collect.
You must contact OpenAI to get the API Key.
Workspace ID—enter the ChatGPT Enterprise WorkspaceID to which you want to connect.

Configuring the data source

Each data source used in Collect has different search criteria on the Collection Details step when creating a collection job. The criteria needs to be configured next.

Relativity collects ChatGPT Enterprise chat data in Relativity's short message format (RSMF). For more information, see The Relativity Short Message Format (RSMF).

Add search criteria to collect specific data. To configure the data sources, complete the following fields:

Select and unselected tabs—choose the data sources to collect from by moving unselected data sources to the selected list.
Field—choose the field to filter on within the data source.
You must set a Start Date to complete the Data Sources step.
Operator—choose an operator. For example, equals or greater than.
Value—enter a value to find in the selected field.

The following table lists the filter criteria supported for ChatGPT Enterprise collections.

Criteria	Operators	Description	Example
Start Date	Greater Than or Equals	When you use the Start Date property in a query, the search returns prompts and responses that exist the day of and after the entered date.	When you search a Start Date of 1/1/2024, Relativity collects all prompts and responses from that date to today.

Temporary, or ephemeral, chats will also be returned if they occurred within the 30-day retention window and meet your search criteria.