Pega GenAI architecture
Pega GenAI™ acts as a secure and managed bridge between Pega Infinity™ enterprise workflow features and third-party large language models (LLMs). Pega GenAI is a suite of Pega Platform™ features that augment design time (developer productivity) and run time (employee and customer experience) by using LLMs through a governed, cloud-hosted control plane.
Pega GenAI aims to be:
- Composable at the Application Layer (Rules and Case Life Cycles drive usage).
- Provider agnostic at the model layer (Azure OpenAI, AWS Bedrock, Google Vertex, and others).
- Governed and secured by using a gateway service for centralized policy enforcement, telemetry, and masking.
- Observable with usage tracking and audit trails for enterprise compliance.
Rather than embedding a single model in Pega Platform, the architecture uses a layered design for flexibility, security, and abstraction. This approach enables Channels (for example, customer service Portals, email bots, and voice systems) to use generative AI without requiring knowledge of the underlying model provider, such as AWS, Google, or OpenAI. All processing occurs securely inside Pega Infinity before any request is sent externally.
Pega GenAI integration is center out: Channels and the UI act as entry points, while orchestration occurs in Pega Infinity, which invokes Pega GenAI services through standardized Rule Types such as Pega Connect GenAI, Pega GenAI Coach™, and Pega GenAI Agent. This Center-out® approach supports strong governance, data minimization, and reusability.
Architectural components breakdown
The architecture includes four layers that flow from interaction to intelligence and back.
- Interaction layer
- Orchestration layer
- Abstraction and processing layer
- LLM models layer
Interaction layer
This layer represents the Channel layer, where the need for generative AI is identified. It captures user or system-initiated input and triggers a Pega GenAI request.
Orchestration layer
The core application receives triggers from the channels and prepares the request using internal rule structures. It determines that a Pega GenAI task is needed and invokes the appropriate components. These components include Connect GenAI, GenAI Coach, and Pega GenAI Agent Rules, which organize data before passing it to Pega GenAI services.
Abstraction and processing layer
This specialized layer within the Pega Cloud® environment manages secure interaction with external AI models. It abstracts model behavior, formats data, and enforces governance.
Key components include:
- Assistant service: Manages the Pega GenAI workflow.
- Payload translation: Formats Pega requests into LLM-readable prompts and converts model responses back into structured workflow output.
- Usage tracking: Monitors request and token volume for auditing, billing, and governance.
- Gateway proxy: The secure boundary that routes sanitized requests to external providers.
LLM models layer
This layer contains third-party foundational models hosted outside the Pega environment. The models perform summarization, generation, extraction, or other cognitive tasks. The model receives the prompt, generates the completion, and returns the response to the Gateway proxy. The architecture is provider agnostic, supporting AWS Bedrock, Google Cloud, and OpenAI models.
The following diagram shows the high‑level Pega GenAI architecture. The diagram illustrates how requests flow from Pega Channels through Pega Infinity and Pega GenAI services in Pega Cloud before the system securely routes the requests to external LLM providers.
A thorough understanding of Pega GenAI architecture is essential for a Lead System Architect (LSA). This knowledge is fundamental to designing and delivering modern, intelligent, and responsible enterprise solutions. As an LSA, you ensure that applications are scalable, secure, and maintainable. Understanding this architecture enables you to make informed design decisions, such as selecting the appropriate Pega GenAI features for a business problem and selecting the right LLM provider while considering cost, performance, and compliance.
This knowledge is also critical for implementing effective governance and security controls. When you understand how Pega GenAI services handle payload translation and protect data through the Gateway proxy, you can design solutions that use generative AI without exposing sensitive enterprise data. This understanding helps you architect applications that adapt to future needs and avoid vendor lock‑in. As a result, the solutions that you create remain flexible as the generative AI landscape evolves. Mastering this architecture is essential for any LSA who is building the next generation of intelligent Pega applications.
Check your knowledge with the following interaction:
This Topic is available in the following Module:
If you are having problems with your training, please review the Pega Academy Support FAQs.
Want to help us improve this content?