Skip to content

Key terms and concepts

This page defines key terms used throughout the OPAQUE documentation. Terms are grouped by concept to help explain how the platform is structured and how its core components relate to each other.

Platform structure

  • Organization: The top-level unit for user management in Opaque, representing one or more users. Each user belongs to exactly one organization. See also role-based access control (RBAC).
  • Workspace: An isolated environment where users collaborate on shared data and compute. Opaque supports two workspace types: agentic AI workspaces (for building and running AI workflows with agents and LLMs) and analytics and ML workspaces (for running jobs over encrypted data). Each workspace enforces strict access controls and data boundaries.
  • Workflow: In OPAQUE, a persistent, callable service built in an agentic AI workspace. A workflow defines how inputs flow through a sequence of steps—such as retrieving context, calling models, applying logic, and producing outputs—and can be invoked repeatedly through the OPAQUE SDK or API.
  • Node: A step within an agentic workflow. Nodes represent individual capabilities—such as calling a large language model, retrieving data through a connector, applying transformations, or running utilities like redaction. Nodes are connected to define the end-to-end behavior of a workflow.
  • Agent: A specialized type of node that performs goal-directed reasoning within an agentic workflow. Not all nodes are agents; many nodes represent model calls, retrieval steps, or utilities. Agents are used where autonomous decision-making or multi-step reasoning is required.
  • Workload: A workload is a general term for a unit of computation in OPAQUE. Both agentic workflows and analytics jobs are workloads, but they differ in execution model and lifecycle.
  • Job: A batch workload that runs to completion in an analytics and ML workspace. Jobs are typically written in Python (PySpark) or SQL and operate on protected datasets. Unlike workflows, jobs are not persistent services and are executed as discrete runs.
  • Dataset: An encrypted dataset added to a workspace. All datasets in Opaque remain encrypted at rest, in transit, and in use, ensuring full privacy throughout their lifecycle. The user who provisions a dataset controls access, but datasets can be shared with workspaces based on assigned permissions. In multi-member workspaces, data-processing jobs require the approval from all members before execution; in single-member workspaces, approval is not needed.
  • Synthetic data: Artificially generated data that mimics the structure and statistical properties of real datasets. Used in analytics and ML workspaces to develop and test jobs without exposing sensitive information. This enables users to refine workflows, test queries, and conduct analysis securely, supporting compliance with privacy regulations.

Access and governance

  • Role-based access control (RBAC): A security model that restricts access based on user roles. In Opaque, these roles include:
    • Organization admin: Manages user roles and organization-wide settings.
    • Workspace admin: Creates workspaces, invites members, and manages workspace access.
    • Workspace member: Works within assigned workspaces, provisioning data, running jobs, and collaborating based on workspace policies.

Security and trust

  • Confidential computing: A security model that protects data even while it’s being processed, using hardware-based environments called trusted execution environments (TEEs). See also trusted execution environments.
  • Confidential AI: An extension of confidential computing that secures entire AI workflows—including data, models, and code—while enabling policy enforcement, collaboration, and verifiable auditability.
  • Opaque trusted computing environment (O-TCB): A verifiable trusted execution boundary within the Opaque platform that provides end-to-end assurance of data and workload confidentiality, integrity, and policy enforcement. It consists primarily of the data plane Kubernetes cluster, where the workload manager and workload processor operate on AMD SEV-SNP–enabled nodes under a hardware root of trust. The O-TCB ensures that all sensitive workloads execute in a cryptographically verified environment, isolated from the broader Azure infrastructure. Services external to this boundary—such as storage or DNS—interact with the O-TCB through encrypted channels and policy-governed interfaces.
  • Trusted execution environment (TEE): A secure, hardware-based enclave that processes encrypted data. TEEs decrypt data only inside the enclave during computation, preventing access by external entities (e.g., cloud providers, administrators, or attackers). Opaque uses TEEs to maintain data privacy throughout computation.
  • Remote attestation: A cryptographic process that verifies a trusted execution environment (TEE) is genuine and running approved code. This allows users to safely share encryption keys and run jobs only in verified, secure environments.