Set up OpenAI connector

ConductorOne provides identity governance and just-in-time provisioning for OpenAI. Integrate your OpenAI instance with ConductorOne to run user access reviews (UARs) and enable just-in-time access requests.

This connector is in beta. This means it’s undergoing ongoing testing and development while we gather feedback, validate functionality, and improve stability. Beta connectors are generally stable, but they may have limited feature support, incomplete error handling, or occasional issues.

We recommend closely monitoring workflows that use this connector and contacting our Support team with any issues or feedback.

Capabilities

Resource	Sync	Provision
Account
Invitation
Role
Project
API key

Note that custom roles are not currently supported. The connector currently supports only predefined static roles:

• Organization roles: Owner, Reader
• Project roles: Owner, Member, Viewer

Additional functionality:

Supports automatic account provisioning and deprovisioning

SCIM provisioning notes

OpenAI supports SCIM-based directory sync for automating user provisioning and deprovisioning. However, there are important limitations to be aware of when planning your provisioning strategy.

Availability

SCIM is only available on certain OpenAI plans:

• API Platform: Custom or Unlimited billing plans
• ChatGPT: Enterprise or EDU plans

SCIM is not available on the Business plan. If you are on a plan that does not support SCIM, you must manage users manually or through the OpenAI Admin API.

Configuring SCIM

SCIM is not configured through the OpenAI API. Instead, organization owners enable directory sync through the Identity settings in the OpenAI Platform dashboard. This opens a setup flow through the WorkOS portal, where you connect your identity provider (IdP).

Supported identity providers include Okta, Entra ID (Azure AD), Google Workspace, PingFederate, OneLogin, and Rippling, among others.

Limitation: no SCIM Groups support on the API Platform

OpenAI’s SCIM integration on the API Platform does not currently support SCIM Groups. This means:

• Users provisioned via SCIM are automatically invited to the organization, but they are not assigned to any project.
• There is no way to use SCIM to automatically map IdP groups to OpenAI projects or roles.
• Each user provisioned via SCIM must be separately assigned to projects after they join the organization.

SCIM Groups are supported for ChatGPT Enterprise workspaces but not for the API Platform. If you use both products, be aware that group sync behavior differs between them.

Workaround: use the Admin API for project assignments

Because SCIM cannot assign users to projects, you can use the OpenAI Admin API’s Project Users endpoints to manage project membership after a user is provisioned via SCIM.

The key endpoints are:

• Add user to project: POST /v1/organization/projects/{project_id}/users — requires user_id and role (member or owner)
• Modify project user role: POST /v1/organization/projects/{project_id}/users/{user_id} — update a user’s project role
• Remove user from project: DELETE /v1/organization/projects/{project_id}/users/{user_id}

The ConductorOne OpenAI connector uses the Admin API to manage project memberships. When you configure ConductorOne to provision access to OpenAI projects, ConductorOne handles these API calls for you, filling the gap that SCIM leaves for project-level access.

What SCIM provides vs. what requires the Admin API

Capability	SCIM	Admin API (via ConductorOne)
Provision user into organization
Deprovision user from organization
Assign user to project
Assign project role
Assign organization role
Map IdP groups to projects

Neither SCIM nor the Admin API currently supports automatically mapping IdP groups to OpenAI projects. To assign users to projects, use ConductorOne’s provisioning capabilities with the OpenAI connector, which calls the Admin API on your behalf.

Gather OpenAI configuration information

Configuring the connector requires you to pass in information from OpenAI. Gather these configuration details before you move on.

Here’s the information you’ll need:

• OpenAI Admin API Key — This must be an Admin API key created with All permissions selected. The connector requires full admin permissions to manage users, projects, and invitations. Keys created with “Restricted” permissions will cause the connector to error.

To create the Admin API key:

1. In the OpenAI platform, navigate to Settings > API keys in the Organization section.
2. Click Create new admin key.
3. When prompted to set permissions, select All. Do not select “Restricted”, as the connector requires access to all organization management endpoints.
4. Copy and save the generated key.

See the OpenAI docs for more information: Admin API Keys documentation

Configure the OpenAI connector

To complete this task, you’ll need:

• The Connector Administrator or Super Administrator role in ConductorOne
• Access to the set of OpenAI configuration information gathered by following the instructions above

Cloud-hosted

Follow these instructions to use a built-in, no-code connector hosted by ConductorOne.

1. In ConductorOne, navigate to Integrations > Connectors and click Add connector.

2. Search for OpenAI and click Add.

3. Choose how to set up the new OpenAI connector:

   • Add the connector to a currently unmanaged app (select from the list of apps that were discovered in your identity, SSO, or federation provider that aren’t yet managed with ConductorOne)
   • Add the connector to a managed app (select from the list of existing managed apps)
   • Create a new managed app
4. Set the owner for this connector. You can manage the connector yourself, or choose someone else from the list of ConductorOne users. Setting multiple owners is allowed. If you choose someone else, ConductorOne will notify the new connector owner by email that their help is needed to complete the setup process.

5. Click Next.

6. Find the Settings area of the page and click Edit.

7. Enter the configuration information from the previous section.

8. Click Save.

9. The connector’s label changes to Syncing, followed by Connected. You can view the logs to ensure that information is syncing.

That’s it! Your OpenAI connector is now pulling access data into ConductorOne.

Self-hosted

Follow these instructions to use the OpenAI connector, hosted and run in your own environment.

When running in service mode on Kubernetes, a self-hosted connector maintains an ongoing connection with ConductorOne, automatically syncing and uploading data at regular intervals. This data is immediately available in the ConductorOne UI for access reviews and access requests.

Resources

• Official download center: For stable binaries (Windows/Linux/macOS) and container images.

Step 1: Set up a new OpenAI connector

1. In ConductorOne, navigate to Integrations > Connectors > Add connector.

2. Search for Baton and click Add.

3. Choose how to set up the new OpenAI connector:

   • Add the connector to a currently unmanaged app (select from the list of apps that were discovered in your identity, SSO, or federation provider that aren’t yet managed with ConductorOne)
   • Add the connector to a managed app (select from the list of existing managed apps)
   • Create a new managed app
4. Set the owner for this connector. You can manage the connector yourself, or choose someone else from the list of ConductorOne users. Setting multiple owners is allowed. If you choose someone else, ConductorOne will notify the new connector owner by email that their help is needed to complete the setup process.

5. Click Next.

6. In the Settings area of the page, click Edit.

7. Click Rotate to generate a new Client ID and Secret. Carefully copy and save these credentials. We’ll use them in Step 2.

Step 2: Create Kubernetes configuration files

Create two Kubernetes manifest files for your OpenAI connector deployment:

Secrets configuration

# baton-openai-secrets.yaml
apiVersion: v1
kind: Secret
metadata:
  name: baton-openai-secrets
type: Opaque
stringData:
  # ConductorOne credentials
  BATON_CLIENT_ID: <ConductorOne client ID>
  BATON_CLIENT_SECRET: <ConductorOne client secret>

  # OpenAI config
  BATON_OPENAI_ADMIN_KEY: <OpenAI Admin API Key with All permissions>

  # Optional: include if you want ConductorOne to provision access using this connector
  BATON_PROVISIONING: true

See the connector’s README or run --help to see all available configuration flags and environment variables.

Deployment configuration

# baton-openai.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: baton-openai
  labels:
    app: baton-openai
spec:
  selector:
    matchLabels:
      app: baton-openai
  template:
    metadata:
      labels:
        app: baton-openai
        baton: true
        baton-app: openai
    spec:
      containers:
      - name: baton-openai
        image: ghcr.io/conductorone/baton-openai:latest
        imagePullPolicy: IfNotPresent
        env:
        - name: BATON_HOST_ID
          value: baton-openai
        envFrom:
        - secretRef:
            name: baton-openai-secrets

Step 3: Deploy the connector

1. Create a namespace in which to run ConductorOne connectors (if desired), then apply the secret config and deployment config files.

2. Check that the connector data uploaded correctly. In ConductorOne, click Apps. On the Managed apps tab, locate and click the name of the application you added the OpenAI connector to. OpenAI data should be found on the Entitlements and Accounts tabs.

That’s it! Your OpenAI connector is now pulling access data into ConductorOne.