Skip to main content

Security

This guide covers MatrixHub's current project-level access isolation, personal and robot credentials, and role-based access control.

🔒 1. Multi-Tenant Project Isolation​

MatrixHub uses a Project as the boundary for model resources and permissions.

  • Project Visibility: All users can view and pull models from public projects. Private projects are available only to project members and platform administrators. Pushing and deleting still require the appropriate project role.
  • Personal Access Tokens: Represent the owning user, are not bound to a single project, and use that user's access scope and role permissions in each project.
  • Robot Accounts: Project permissions can apply to all projects or selected projects. In selected-project mode, the configured project permissions apply only within those projects; existing read-only access to public projects is unaffected.

Configuration references: Create and Delete Projects, Personal Access Tokens, and Robot Accounts.

🛡ī¸ 2. Fine-Grained Role-Based Access Control (RBAC)​

MatrixHub separates platform-level roles from project-level roles. A platform administrator (admin) has global platform permissions, while access to models within a project is controlled by three project roles:

RolePermission Scope
Project ViewerView the project and member list, and view and pull models. Cannot push or delete models or manage project members.
Project EditorIncludes Project Viewer permissions and can push models. Cannot delete models or manage project members.
Project AdminIncludes Project Editor permissions and can add or remove members, change member roles, update project settings, and delete models and the project.
Platform Admin (admin)Can use platform administration features and perform all operations within every project.

Every user who can sign in to MatrixHub can create a project and automatically becomes the Project Admin of that project. Members cannot remove themselves from the current project.

Configuration reference: Project Members and Roles.

🚧 3. Planned Security Capabilities​

The following capabilities are not available in the current release and are listed in the MatrixHub Roadmap:

  • LDAP, OIDC, or SSO integration
  • Audit logging
  • Malicious model content scanning, model signing, and signature verification