Skip to content

Multi-tenancy model

Opsta AI Gateway is multi-tenant by design. One platform serves many organizations, each fully isolated, and within each organization a clear hierarchy controls configuration, budgets, and access.

The hierarchy

  • Organization — the isolation, billing, and SSO boundary. One enterprise customer. Each organization connects its own identity provider and cannot see another's data, config, or telemetry.
  • Project — owns a routing configuration, providers, guardrails, budgets, and API keys. An organization has many projects (for example, one per product or environment).
  • Group — a team within a project, usually mapped from an identity-provider group. Used for access rollups and budget aggregation.
  • User — a member who signs in and/or calls the gateway.

Every API key, budget, limit, usage record, and metric is keyed by the full tuple organization.project.user — so attribution and isolation are exact, never a flat shared key.

Roles and access (RBAC)

RoleCan do
Platform adminManage every organization; set global model pricing; read the cross-organization audit log; configure platform login methods.
Org adminManage one organization — members, projects, providers, routing, budgets, guardrails, MCP servers, and the organization's identity provider.
MemberUse the gateway: issue and manage their own API keys, view their usage and budget, and review their blocked requests.

Members are global identities; membership in an organization carries a role, and a user can belong to more than one organization. See Organizations & members.

Hierarchical budgets

Budgets cascade down the hierarchy, and the tightest cap always wins — a user can never spend more than their group, project, or organization allows, even if their own cap is higher.

This lets platform owners set a hard ceiling at the top while delegating finer limits downward. See Budgets & limits.

Isolation in observability

Each organization gets its own isolated dashboards and metrics tenancy — one customer's usage and telemetry are never visible to another. See Observability.

Enterprise AI governance, on infrastructure you own.