The MPAI as a Service (MPAI‑MAS) V1.0 standard enables remote access to AI processing capabilities, for example hosted by a cloud service. It allows a Remote Client Application to request the execution of an AI Module (AIMs) on an AI Framework. The MPAI AI Framework (MPAI‑AIF) standard offers the necessary functionality in support of MPAI-MAS, thus enhancing interoperability and modularity.

The standard enables MPAI AI processing to be offered as a service, allowing applications to dynamically access and use AI components functioning on remote systems.

The Reference Architecture

MPAI‑MAS is based on a service-oriented architecture where a Remote Client Application interacts with an AI Framework instance through a standard HTTPS interface.

The system includes:

  • A Remote Client Application (RCA) requesting AI processing services.
  • A Service Interface exposing AIF functionality through standard APIs.
  • A Service Controller Instance (SCI) managing AIM execution.
  • The MPAI Store providing AIM descriptions and components.

Components interact through well-defined interfaces, enabling independent development and deployment while maintaining interoperability.

Main Functions of MPAI as a Service

The MPAI‑MAS standard will enable the following essential functions:

  • Remote Execution of AIMs – Allows a client to request execution of simple or composite AIMs.
  • Dynamic Configuration – Enables instantiation and subsequent control of AIMs at runtime.
  • Data Exchange – Supports dynamic delivery of input data and retrieval of output results.
  • Life-cycle Management – Controls creation, execution, and termination of AIMs.
  • Application Selection – Searches AIM descriptions from the MPAI Store.
  • Resource Access –  Downloads AIMs from distributed repositories.

These functions allow applications to access AI capabilities without requiring local implementation of AIMs.

Operation Model

MPAI‑MAS separates control and data planes to ensure efficient and secure operation:

  • The control plane manages AI Module life-cycle (creation, execution, termination).
  • The data plane handles dynamic application data exchange with AIMs.

A Remote Client Application can:

  • Request creation of a Service Controller Instance.
  • Launch and control AIMs.
  • Send required input data to the AIMs.
  • Receive output results synchronously or asynchronously.
  • Terminate running AIMs once processing is completed.

This model supports scalable and flexible deployment of MPAI AI services.

Powered by the MPAI AI Framework

MPAI‑MAS relies on the MPAI Artificial Intelligence Framework (MPAI‑AIF V3.0 Basic Profile), which provides:

  • A modular and interoperable execution environment.
  • Dynamic configuration and orchestration of AIMs.
  • Platform-independent implementation.
  • Operation in Zero‑Trust local or distributed environments.

The AIF enables standardised execution of AIMs, making them accessible online as reusable services.

Open and Flexible Design

MPAI‑MAS follows key design principles:

  • Interoperable service-based access to AI applications.
  • Modular and reusable AI components.
  • Standard interfaces for remote interaction.
  • Independence from specific cloud or infrastructure technologies.

This enables deployment across different environments, including cloud, edge, and hybrid systems.

Benefits

The MPAI‑MAS standard will support:

  • Service Providers to offer AI applications as scalable cloud services.
  • Developers to integrate AI capabilities without needing to control underlying models.
  • Integrators to compose applications from distributed AI components.
  • Enterprises to deploy flexible and interoperable AI solutions.
  • Users to access advanced AI functionalities on demand.

A New Paradigm for AI Deployment

MPAI‑MAS promotes a shift from locally deployed AI systems to:

  • On-demand interoperable AI services accessible over networks.
  • Distributed AI processing across multiple environments.
  • Reusable AIMs delivered as services.
  • Scalable and dynamically configurable AI systems.

This approach enables rapid development of AI-driven applications.

Conclusion

MPAI‑MAS V1.0 provides a complete and interoperable framework for delivering AI applications as services that:

  • Enable remote execution of AIMs.
  • Support dynamic configuration and control of AI processing.
  • Guarantee interoperability across implementations.
  • Allow scalable deployment in cloud and other distributed environments.

The standard enables a new generation of AI applications based on service-oriented architectures and reusable AI components.