All Products
Search
Document Center

Alibaba Cloud Model Studio:Permission management

Last Updated:Nov 05, 2026

Alibaba Cloud Model Studio provides granular access control at the console and model levels to support complex organizational structures spanning multiple regions.

Alibaba Cloud Model Studio identity management

A single workspace is the smallest management unit for fine-grained permission management (models and users) and Alibaba Cloud cost allocation.

Model Studio workspace permission management is based on three roles:

  1. Super administrator: Can manage user permissions, workspace-available models, workspace model rate limiting, and API keys in a unified, cross-workspace manner.
  2. Workspace administrator: Responsible only for user permissions and resource management within a specific workspace.
  3. Regular user: Uses resources based on the assigned permissions.

Workspace permission

Super administrator (with the AliyunBailianFullAccess system policy)

Workspace administrator

Regular user

Allow specific model calls & rate limiting

Supported

Not supported

Not supported

Allow specific model fine-tuning

Supported

Not supported

Not supported

User management

Supported

Supported

Not supported

User page access management

Supported

Supported

Not supported

API key management

Supported

Supported

Not supported

Access/use authorized workspaces, pages, and resources

Supported

Supported

Supported

OpenAPI permissions

Not supported

Not supported

Not supported

Super administrator

It includes the following two types of accounts:

  • An Alibaba Cloud primary account. You can view it in the upper-right corner of the Model Studio console:

    Click the profile icon to expand the account drop-down panel. The entry with the Primary account label is the super administrator account.

  • A RAM user (account) with the AliyunBailianFullAccess (Model Studio administrator) system policy. In the upper-right corner of the Model Studio console, click the profile icon to expand the account drop-down panel and view the information of the currently logged-on RAM user account. This RAM user can use the global management menu of Model Studio (Singapore | Virginia | Beijing | China (Hong Kong) | Frankfurt) to grant almost all permissions in any region and any workspace to any RAM user (including themselves). (Only OpenAPI permissions can be added by the Alibaba Cloud primary account.)

    A RAM user is a sub-account created by an Alibaba Cloud primary account. It is used to securely allocate cloud resources and permissions to team members.

Super administrators can use the global management menu of Model Studio (Singapore | Virginia | Beijing | China (Hong Kong) | Frankfurt) to manage multiple workspaces. The features include:

  1. Create workspaces and manage workspace names.
  2. Manage models and model rate limiting for all workspaces.
  3. Manage accounts (users) for all workspaces.
  4. Manage all API keys.

NoteTo enable features such as the AI guardrails service (available only in the China (Beijing) region) and model monitoring, we recommend that you use the Alibaba Cloud primary account to perform a one-time authorization and activation in the console.

Workspace administrator

A workspace administrator refers to an Alibaba Cloud RAM user who has access to the Permission Management page of a specific workspace. The user can manage the workspace through that page.

The Administrator permission includes the permission to access all pages in the workspace.

In the Edit Permissions dialog box, select the Other tab, select Administrator in the permission name list, and then click OK.

Workspace permission management

Model Studio divides resources and workspaces by geographic region. A single workspace cannot span regions. Even the default workspaces in different regions are different workspaces. Click to go to the global management menu (Singapore | Virginia | Beijing | China (Hong Kong) | Frankfurt).

A workspace in Model Studio is also the smallest management unit for fine-grained permission management. It can manage:

Workspace permission

Super administrator (with the AliyunBailianFullAccess system policy)

Workspace administrator

Regular user

Allow specific model calls & rate limiting

Supported

Not supported

Not supported

Allow specific model fine-tuning

Supported

Not supported

Not supported

User management

Supported

Supported

Not supported

User page access management

Supported

Supported

Not supported

API key management

Supported

Supported

Not supported

Access/use authorized workspaces, pages, and resources

Supported

Supported

Supported

OpenAPI permissions

Not supported

Not supported

Not supported

  • Restrict model calls: Manage whether a model can be called in the workspace (console & API) and set the request rate limit and token limit for the model.

    The default workspace does not support this restriction. All models can be called and cannot be rate-limited.

    In the model list, use the toggle in the Model Calling column to control the model authorization status, and set the request rate limit value, token limit value, and the corresponding time unit in the Current Workspace Limit column (1 second, 2 seconds, 3 seconds, 5 seconds, 10 seconds, 30 seconds, and 60 seconds are supported). If no rate limit is set, the workspace shares the account-level total rate limit quota.

  • Restrict model training: Manage whether a model can be fine-tuned in the workspace (through the console and API) and deployed after fine-tuning.

    The default workspace does not support this restriction. All models that support fine-tuning can be fine-tuned and deployed after fine-tuning is complete.

    On the Model List tab of the workspace, search for and find the target model, and turn the authorization toggle in the Model Training column of the Model Authorization area on or off to control the model training permission.

  • User (account) console permission management: Manage whether a RAM user can use the features of the workspace console and which features of the workspace console the user can use. Console page permissions do not affect the user's model calls through an API key (the permissions of an API key are determined by the model authorization of the workspace it belongs to).

    The Alibaba Cloud primary account does not require this setting and can access all pages of all workspaces.

    In the Edit Permissions dialog box, select the Model tab, find and select the Model Experience - Operations permission item in the permission list. The Experience Center in the left-side navigation pane corresponds to the Model Experience - Operations item in the permission list.

API key permissions

A single API key can belong to only one workspace and one user within one region, and cannot be transferred to another workspace or another user. The functions that an API key can call and its model rate limiting are consistent with the permissions of the workspace it belongs to, and are not affected by user (account) console permission management. You also do not need to create different API keys for different models (such as text-to-text, text-to-image, and speech synthesis).

The status of an API key changes with the operations of the user (account) it belongs to, as follows. Both the primary account and RAM users with API key permission for the workspace can delete API keys.

Note

  • Starting March 25, 2026, all newly created API keys in the China (Beijing) region belong to the primary account.
  • API keys created by a RAM user before March 25, 2026 are bound to that RAM user. If you delete the RAM user in the RAM console, Model Studio also deletes these API keys, and they cannot be recovered. Before deleting a RAM user, check whether the API keys under the user are in use. If you need to retain the calling capability, create new API keys with the primary account or another account and update your business configuration.
  • API keys created on or after March 25, 2026 are not affected by RAM user deletion. After the RAM user is deleted, these API keys are retained and remain available.
ActionAPI keys of the primary accountAPI keys of RAM accounts
Manually delete an API key

Supported (the primary account has this permission by default; RAM users need the API key management permission for the workspace). The key becomes invalid immediately and cannot be recovered.

Supported (the primary account has this permission by default; RAM users need the API key management permission for the workspace). The key becomes invalid immediately and cannot be recovered.

Remove an account from a workspace

—

The key becomes invalid.

The API key becomes effective again after the account is added back to the workspace.

Delete the account/role in theRAM console

—

API keys created before March 25, 2026: deleted with the RAM user and cannot be recovered.

API keys created on or after March 25, 2026: not affected and remain valid.

Set an IP whitelist for the API key

Supported for API keys in all regions. The US (Virginia) region supports IPv4 only.

Supported for API keys in all regions. The US (Virginia) region supports IPv4 only.

Manage API keys: In the left-side navigation pane of the Model Studio console, choose Account Management, find the target RAM user, and click Permission Management to add the API key permission for the user. This grants the RAM user the permissions to create, delete, and view all API keys in the workspace.

In the Edit Permissions dialog box, switch to the Other tab and select API Key.

OpenAPI permissions

By default, RAM users cannot call the Open API for features such as the data, knowledge base, and prompt engineering of Model Studio applications.

To call these APIs, the Alibaba Cloud primary account must add one of the following permissions for the RAM user in the RAM console:

To implement more fine-grained permission control over the knowledge base, you can create a custom policy in the RAM console to grant only the permissions required for the operations.

Knowledge base-related actions: sfm:CreateIndex, sfm:UpdateIndex, sfm:DeleteIndex, sfm:SubmitIndexJob.

Example custom policy:

{
  "Version": "1",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "sfm:CreateIndex",
        "sfm:UpdateIndex",
        "sfm:DeleteIndex",
        "sfm:SubmitIndexJob"
      ],
      "Resource": "*"
    }
  ]
}

After you create the policy, attach it to the RAM user to grant the minimum CRUD permissions for the knowledge base. For the complete list of actions, see RAM authorization.

Apply to production environments

  • Workspace planning strategy
    • By environment (recommended): Create separate workspaces for the development, test, staging, and production environments to achieve strict environment isolation.

      • project-dev-workspace
      • project-test-workspace
      • project-prod-workspace
    • By business line: Create separate workspaces for different business departments in your company (such as marketing, after-sales, and design) to facilitate permission and cost management.

      • marketing-team-workspace
      • customer-team-workspace
  • Rate limiting strategy
    • Allocate the total quota of the primary account proportionally to each workspace, and reserve a portion as a buffer to handle unexpected traffic spikes.

      Example: The total account quota is 1,000 QPM. Allocation plan:

      • project-prod-workspace: 600 QPM (60%)
      • project-test-workspace: 200 QPM (20%)
      • project-dev-workspace: 100 QPM (10%)
      • Reserved buffer: 100 QPM (10%)

Bill viewing and pre-paid permission management

By default, RAM users cannot view Alibaba Cloud bills or purchase Alibaba Cloud pre-paid products. To grant these permissions to a RAM user, add specific permissions for the RAM user in the RAM console.

NoteThe following permissions grant the RAM user the permission to view the bills of all Alibaba Cloud products or purchase all Alibaba Cloud pre-paid products. Grant these permissions with caution.

  1. To view Alibaba Cloud bills, add the AliyunBSSReadOnlyAccess permission for the RAM user.
  2. To purchase Alibaba Cloud pre-paid products, add the AliyunBSSOrderAccess permission for the RAM user.

Common settings

Configure a super administrator

This operation must be performed by an Alibaba Cloud account (primary account) or a RAM user with the AliyunRAMFullAccess system policy.

  1. Go to the RAM console and add the AliyunBailianFullAccess (Model Studio administrator) permission and the AliyunBSSOrderAccess (for purchasing Alibaba Cloud pre-paid products) permission for the RAM user.
  2. After the configuration is complete, you can use the global management menu of Model Studio (Singapore | Virginia | Beijing | China (Hong Kong) | Frankfurt) to grant any permission in any region and any workspace to any RAM user (including yourself), and purchase Model Studio pre-paid products.

Configure a workspace administrator

This operation must be performed by a super administrator or a workspace administrator.

  1. In the left-side navigation pane of the Model Studio console, choose Account Management, find the target RAM user, and click Permission Management to add the Administrator permission for the user.

    In the Edit Permissions dialog box, switch to the Other tab and select Administrator.

Configure model calling permissions

  1. If you do not use the default workspace, ensure that the workspace has the model calling permission enabled for the specific model. (This operation requires a super administrator.)

  2. To call models from the Model Studio console, choose Account Management in the left-side navigation pane of the Model Studio console, find the target RAM user, and click Permission Management to add the following permissions for the user: (This operation requires a super administrator or a workspace administrator.)

    1. The Model Experience - Operations permission, used to call models on the console.
    2. The Batch Inference - Operations permission, used to support the batch inference feature.
    3. The Model Monitoring - Operations permission, used to view the token consumption of model calls and evaluations.
  3. To call models using the Model Studio API, create or assign an API key for the RAM user in the corresponding workspace. For more details, see API key permissions in this topic. (This operation requires a super administrator or a workspace administrator.)

Configure API model fine-tuning permissions

  1. If you do not use the default workspace, ensure that the workspace has the model fine-tuning (training) permission enabled for the specific model. (This operation requires a super administrator.)
  2. Create or assign an API key for the RAM user in the corresponding workspace. For more details, see API key permissions in this topic. (This operation requires a super administrator or a workspace administrator.)

FAQ

1. How do I obtain the workspace ID?

See Obtain the Workspace ID in application development.

2. How do I call a model from a sub-workspace?

No special settings are required. You can use the API key of the sub-workspace.

3. How do I use applications in a specific workspace?

To manage and call applications in a specific workspace by using the API, you must set the APP ID and Workspace ID at the same time.

4. What should I do if AliyunBailianFullAccess is configured but I still cannot manage workspaces?

RAM permissions (such as AliyunBailianFullAccess) and the workspace-level permissions inside the Model Studio platform are two independent permission systems. RAM permissions only control access to the Model Studio console, whereas the management permissions in a workspace (such as editing the workspace and managing members) must be granted separately by the primary account or a super administrator in the Model Studio console.

If a sub-account already has the AliyunBailianFullAccess permission but still receives a message indicating that it has no permission to operate workspaces, perform the following steps:

  1. Log on to the Model Studio console with the primary account.
  2. Click the settings icon in the upper-right corner and choose Account Management in the left-side navigation pane.
  3. Find the sub-account that requires authorization and click Permission Management.
  4. Add the management permission for the corresponding workspace to the sub-account.

5. How do I delete a workspace?

Go to the Alibaba Cloud Model Studio console, select the target region in the upper-right corner of the page, go to the Workspace Management page, and select Delete in the Actions column; or call the DeleteWorkspace API (DELETE /modelstudio/workspaces/{workspaceId}). Deleted workspaces cannot be recovered.

Deletion prerequisites: Only workspaces that meet the following conditions can be deleted:

  1. The workspace is not the default workspace.
  2. The workspace is not used to purchase other products, such as Miaobi.
  3. In permission management, the workspace has not been granted to any RAM user or RAM role.
  4. No resources exist in the workspace, such as API keys, model deployments, or knowledge bases.

Clean up dependencies: Before deleting the workspace, clean up the resources of each sub-product in the workspace:

Sub-product

Cleanup content

CCAI conversation analysis AIO

Delete all applications in the workspace and confirm that no data or API keys remain

Multimodal interaction suite

Delete all applications

Tingwu

Search for "Tingwu" in the workbench and delete the corresponding workspace

RAG knowledge base

Delete all services under knowledge management, knowledge retrieval, and knowledge Q&A

Data connection / RAG dataset / Connector

Delete all connectors, uploaded files, tables, and Connector connections

Model training && inference

Delete all datasets, training tasks, my models, and deployments

Observability resources

Disable model monitoring, audit logs, inference log data delivery, and application observability

Evaluation resources

Terminate or delete unfinished model evaluation and application evaluation tasks

If the deletion fails and failReasons is returned, clean up the corresponding resources as prompted and try again.

6. How do I implement more fine-grained permission control over the knowledge base and long-term memory?

You can create a custom policy in the RAM console to grant only the permissions required for the operations.

Knowledge base-related actions: sfm:CreateIndex, sfm:UpdateIndex, sfm:DeleteIndex, sfm:SubmitIndexJob. Long-term memory-related actions: sfm:CreateMemory, sfm:UpdateMemory, sfm:DeleteMemory.

Example custom policy:

{
  "Version": "1",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "sfm:CreateIndex",
        "sfm:UpdateIndex",
        "sfm:DeleteIndex",
        "sfm:SubmitIndexJob",
        "sfm:CreateMemory",
        "sfm:UpdateMemory",
        "sfm:DeleteMemory"
      ],
      "Resource": "*"
    }
  ]
}

After you create the policy, attach it to the RAM user to grant the minimum CRUD permissions for the knowledge base and long-term memory. For the complete list of actions, see RAM authorization.