Study guide

13.3 Pega GenAI Connect Rules, Knowledge Buddy, Agents & Tool Rules

Pegasystems Certified Pega System Architect CSA

Track your progress as you read

Finished sections get a check mark.

1 Chapter 1: Exam Overview & Architecture Foundations

1.1 Certified Pega System Architect Exam Format, Blueprint & Strategy 1.2 Pega Architecture & Enterprise Class Structure (ECS) 1.3 Studio Environments: App Studio, Dev Studio, Admin Studio & Prediction Studio

2 Chapter 2: Case Lifecycle Design & Core Workflows

2.1 Case Types, Stages, Processes, and Steps 2.2 Routing Strategies: Worklists, Work Queues, and Teams 2.3 Service Level Agreements (SLAs) & Urgency Calculations

3 Chapter 3: Advanced Case Management Patterns

3.1 Multi-Step Processes, Parallel Processing & Sub-processes 3.2 Child Cases, Data Propagation & Wait Steps 3.3 Optional Actions, Approvals & Case Resolution 3.4 Duplicate Case Detection, Automation Shapes & Automated Correspondence

4 Chapter 4: Data Modeling & Data Page Architecture

4.1 Data Objects, Property Modes & Page Types 4.2 Data Pages: Scope, Types & Refresh Policies 4.3 Data Transforms & Object Mapping

5 Chapter 5: Business Logic, Calculations & Decisioning

5.1 Declare Expressions & Declarative Network Chaining 5.2 Decision Tables & Decision Trees 5.3 Integration Connectors, Systems of Record & Error Handling

6 Chapter 6: User Interface & User Experience Design

6.1 Traditional UI vs Constellation Architecture 6.2 Views, Dynamic Layouts, Design Templates & Action Sets 6.3 Form Validation, Edit Validate & WCAG Accessibility 6.4 Form Views: Default vs Hierarchical Forms, Field Groups & Primary Fields

7 Chapter 7: Rule Architecture & Rule Resolution

7.1 Rulesets, Versioning & Ruleset Stack Hierarchy 7.2 The Rule Resolution Algorithm 7.3 Circumstancing & Rule Specialization

8 Chapter 8: Security & Access Control

8.1 Authentication, Operator Records & Access Groups 8.2 Role-Based Access Control: Access Roles, ARO & Privileges 8.3 Attribute-Based Access Control (ABAC), Encryption & Audit Trails 8.4 The Security Checklist & Secure Deployment

9 Chapter 9: DevOps, Testing & Application Quality

9.1 Guardrail Compliance, Best Practices & Performance Profiling 9.2 Automated Testing: Pega Unit Test Cases & Scenario Testing 9.3 Packaging, Branching & Deployment Manager CI/CD

10 Chapter 10: Reporting & Operational Analytics

10.1 Report Definitions: Querying, Columns, Summaries & Joins 10.2 Filtering, Parameters & Column Exposure Optimization 10.3 Insights, Visual Dashboards & Business Metrics

11 Chapter 11: Mobility, Digital Channels & Portals

11.1 Mobile Channel Configuration & Offline Processing 11.2 Web Mashup, Digital Channels & Constellation Embed 11.3 Channels, Personas, Portals & Landing Pages

12 Chapter 12: Center-out Architecture, Enterprise Reuse & Agile Delivery

12.1 Center-out Business Architecture & the Case for Pega 12.2 The Situational Layer Cake, Modules & Enterprise Reuse 12.3 Agile Workbench, Workspaces, Branches, Delivery Roles & the Estimator

13 Chapter 13: Pega GenAI for System Architects

13.1 Pega Blueprint: Creating, Importing & Modifying an Application Design 13.2 Pega GenAI Autopilot & Pega GenAI Coach 13.3 Pega GenAI Connect Rules, Knowledge Buddy, Agents & Tool Rules

13.3 Pega GenAI Connect Rules, Knowledge Buddy, Agents & Tool Rules

Key Takeaways Explain This Topic AI

A Pega GenAI Connect Rule integrates generative AI into an application through the Pega GenAI gateway, and can be configured as a Connect Generative AI automation step in App Studio.

Building a Connect Rule starts with a prompt — either custom to the step or an existing prompt authored in Dev Studio — and ends by naming a target field that must also be added to the UI for the response to be visible.

Connect Rule responses come in three structures: Structured - Single maps to one field, Structured - List maps to an embedded data list, and Unstructured is best mapped to a Text (paragraph) field.

Knowledge Buddy answers questions conversationally by retrieving from the organization's own procedural documents and knowledge bases, giving client-specific contextual answers.

A Pega GenAI Agent can run an automation, start a case, call another agent, or launch a user action, and can be given data pages and buddies for context; Tool Rules define those capabilities and supply governance, security, and compliance.

Last updated: September 2026

13.3 Pega GenAI Connect Rules, Knowledge Buddy, Agents & Tool Rules

Blueprint designs the application and Autopilot and Coach assist people. This section covers the three capabilities that put generative AI inside the running application : Connect Rules, Knowledge Buddy, and Agents with their Tool Rules.

1. Pega GenAI Connect Rules

Pega GenAI Connect™ lets you develop custom generative-AI-powered capabilities using the Pega GenAI gateway . It is an add-on compatible with Pega Platform, Pega Customer Service, and Pega Sales Automation.

Typical uses the documentation gives: a Connect Rule can summarize the details of a complex case for a customer service representative brought in to help resolve a complaint, or produce the list of required documents a customer must submit to resolve it.

From Pega Infinity '24 you can configure a Connect Generative AI automation step directly in App Studio — which places it in the same family as the other automation shapes from Section 3.4.

Configuring a Connect Rule

Step 1 — Define the prompt. The prompt defines the request for information sent to the Pega GenAI engine. It can be a custom prompt specific to the step , or an existing prompt created in Dev Studio and reused across steps. Reusing a governed prompt is the maintainable pattern; re-typing a prompt into each step is the same anti-pattern as hard-coding a literal in a view.

Step 2 — Choose the response structure. Pega GenAI responds in one of three structures, and choosing correctly is what makes the response usable:

Response structure What comes back How it maps

Structured - Single A single page with a short paragraph, or a field with a corresponding value The system can map the response directly to a field in the data model

Structured - List A bulleted list of fields and corresponding values The system can map the values directly to the fields of an embedded data list in the data model

Unstructured A long or complex response, such as JSON code Best practice is to map it to a field with a field type of Text (paragraph)

Step 3 — Identify the target field. The system requires you to name a target field to write the response into. Critically, you must also add that target field to the application's UI — otherwise the response is written to the clipboard and the user never sees it. That omission is a favourite exam trap.

Additional Capabilities

Connect Rules include the ability to mask sensitive data during the request and response cycle , and to change the language of responses based on the locale setting of the current user — which ties the capability back to the localization practices in Section 11.3.

2. Pega GenAI Knowledge Buddy

Pega GenAI Knowledge Buddy™ gives employees, agents, and sellers instant answers with real-time guidance and conversational knowledge through a conversation interface. It combines Pega GenAI with information retrieved from the organization's own procedural documents and knowledge bases .

The phrase the blueprint uses is client-specific contextual knowledge and support , and that is precisely the differentiator. A general-purpose model knows general things; a Buddy knows this company's standard operating procedures, policy manuals, and product documentation, because those are the sources it retrieves from. Knowledge Buddy excels at providing client-specific contextual information and answers while a user is learning, developing, or working, and it can be enabled for an intelligent virtual assistant (IVA) so that customers, not only employees, benefit.

Positioning it against its neighbours:

Knowledge Buddy answers questions from your documents. Ask it what the flood-zone underwriting rule is, and it retrieves from your underwriting manual.

Coach advises on the task at hand. It works from the case context and the instructions an architect configured.

A Connect Rule performs a defined generative operation inside a step and writes the result into a case field.

3. Pega GenAI Agents & Tool Rules

Agent and Tool Rules automate case processing and provide conversational support for run-time users. Agents offer automated actions and run-time guidance that accelerate case processing.

What an Agent Can Do

You configure an agent to perform various actions:

Run an automation.

Start a case.

Call another agent.

Launch a user action.

You can also list data pages and buddies to provide more context and data to the agent, and you can extend the agent's range by connecting it to agent-to-agent or MCP protocols .

Tool Rules

Tool rules define the capabilities an agent uses to complete work , and they are the governance layer: Tool Rules provide governance, security, and compliance for what an agent is permitted to do. An agent without carefully scoped tools is an agent with undefined authority, which is the concern any enterprise raises first.

By providing detailed descriptions and guardrails for the agent, you ensure it responds precisely to the user's question rather than improvising.

Run-time Behaviour

At run time, users converse with agents in a chat, using natural language . The agent responds to queries and performs its defined actions, accelerating case processing. During a conversation, agents can also process and extract data from uploaded documents , removing manual analysis time.

The business effect Pega claims is straightforward: agents increase automation in processes, eliminate repetitive tasks, and reduce the risk of human error in case processing.

4. Choosing the Right Capability

Requirement Correct capability

"Summarize this case into a field on the review screen" Connect Rule (Structured - Single or Unstructured → Text paragraph), with the target field added to the UI

"Extract the five required documents into a repeating list on the case" Connect Rule with a Structured - List response mapped to an embedded data list

"Answer policy questions from our underwriting manual" Knowledge Buddy

"Let the user ask, in chat, to start a related case and run an automation" Agent with the appropriate Tool Rules

  • "Guide the user through the next best step on this case" Coach
  • "Suggest fields and a life cycle while we build" Autopilot
  • "Design the application before we have an environment" Blueprint

Memorizing that table is worth more exam marks than any amount of general AI knowledge, because the Pega GenAI domain is tested as capability selection, not as machine-learning theory.

Loading diagram... Generative AI Inside the Running Application