OpenAI
Sync OpenAI organization and ChatGPT Enterprise compliance data. This connector syncs organization members and pending invites from the OpenAI Admin API. It also syncs daily completion usage, aggregated by project, user and model. ChatGPT workspace user rosters include current and historical data. For Enterprise workspaces, it also syncs compliance conversation log metadata and message-level events from JSONL export files. You need an OpenAI organization API key with admin access. ChatGPT Enterprise compliance streams also need a compliance API token and workspace ID. Usage and compliance log streams support incremental sync with configurable start and end dates (YYYY-MM-DD).
The connection check runs only the users stream using the OpenAI Admin API. A valid organization API key can pass this check when the compliance token or workspace ID is missing or invalid. Any related errors appear when compliance streams run during a sync.
Set up the connector
API Platform (Admin API): required for users, invites and completions
- Organization ID: In OpenAI Platform organization settings, copy your organization ID, such as
org-…. Enter it as OpenAI-Organization (openai_organization). - Admin API key: An organization owner creates an Admin API key under Organization settings → Admin keys. Use a key with the scopes needed for members, invites and usage. See the Admin APIs guide. Enter the secret as OpenAI-token (
openai_token).
These credentials call api.openai.com for organization users, invites and completion usage. They’re separate from ChatGPT workspace Admin keys.
ChatGPT Enterprise compliance: required only for compliance streams
Compliance streams (conversation_messages, conversation_messages_log, current_users, historical_users) require a ChatGPT Enterprise or Edu workspace. Standard API Platform access isn’t enough.
- Compliance API token: In Admin Console, a workspace owner or admin creates a workspace-scoped Admin key. It needs the compliance permissions required by your integration, such as conversation messages and workspace user read access. See Managing Admin keys in Admin Console. Enter the key as OpenAI-compliance-token (
openai_compliance_token). Only workspace owners can grant broad compliance or conversation-message access. - Workspace ID: In Admin Console, open the target workspace. Copy its workspace ID from the URL or workspace settings. Enter it as Workspace ID (
workspace_id). Compliance API requests usehttps://api.chatgpt.com/v1with this ID in the path.
If you only need organization admin data, leave the compliance token and workspace ID blank. Then disable the compliance streams in the connection schema.
Configure the connector
| Input | Type | Description | Default Value |
|---|---|---|---|
end_time | string (date, YYYY-MM-DD) | End Time. The end time for querying the API in YYYY-MM-DD (defaults to today if omitted). Because the end date is treated as midnight UTC on that day, today’s data is not included until a sync runs after that calendar day ends (typically the next day). | |
start_time | string (date, YYYY-MM-DD) | Start time. The start time for querying the API in YYYY-MM-DD | |
openai_token | string | OpenAI-token. The API key token used for authorizing API requests | |
workspace_id | string | Workspace ID. Required for compliance streams; not validated during connection check. | |
openai_organization | string | OpenAI-Organization. The organization (organization_id) associated with the request | |
openai_compliance_token | string | OpenAI-compliance-token. Required for compliance streams; not validated during connection check (only users is checked). |
Streams
| Stream Name | Primary Key | Pagination | Supports Full Sync | Supports Incremental |
|---|---|---|---|---|
| users | id | DefaultPaginator | ✅ | ❌ |
| conversation_messages | id | DefaultPaginator | ✅ | ✅ |
| conversation_messages_log | event_id.snapshot_date | No pagination | ✅ | ✅ |
| invites | id | DefaultPaginator | ✅ | ❌ |
| completions | snapshot_date.project_id.user_id.model | DefaultPaginator | ✅ | ✅ |
| current_users | id | DefaultPaginator | ✅ | ❌ |
| historical_users | id | DefaultPaginator | ✅ | ❌ |
conversation_messages_log
Each JSONL line is flattened for analytics. actor and principal become the actor_* and principal_* columns. Message text appears in message_content_value, while conversation metadata includes conversation_title. message.author is split into author_type, author_client_type, author_skills_used and author_tools_used. Skills, tools and non-text content are stringified, so nested or tuple-shaped API values have clean types. The raw nested message, conversation, actor and principal objects aren’t emitted.