Skip to Content

Reference architecture for building advanced AI agents with Jitterbit Harmony

Introduction

This page provides an overview of the reference architecture for an advanced AI-powered automation agent, capable of understanding inputs, reasoning with memory and context, and executing real-world operations across enterprise and SaaS systems via connectors and tools.

Reference architecture

Advanced AI agent system diagram

① AI agent core and workflow

At the center of the system design is the agent core. The AI agent core acts as the "brain" of the system and processes inputs (from chat, SaaS apps, websites, microservices, etc.) through input processing and task/action handling.

The AI agent core connects with a large language model (LLM), which powers reasoning, decision-making, and conversation.

② Memory and context

The LLM is supported by different forms of memory and knowledge, allowing the AI agent to recall past interactions, retrieve knowledge dynamically, and maintain continuity in conversations.

  • Context: Immediate session-based knowledge.
  • Short-term memory: Temporary working memory for recent interactions.
  • Long-term memory: More persistent knowledge across sessions.
  • Knowledge base: Provides external retrieval and enrichment of information.

③ Tools and integrations

Based on the decisions the LLM makes, the AI agent core invokes tools (integrations) to perform actions and execute operational tasks across enterprise and SaaS ecosystems.

These tools connect either via the MCP Client connector to an MCP server, or via connectivity to integrations and APIs within or outside the Harmony platform, including these:

  • Enterprise systems
  • SaaS / ERP / CRM platforms
  • Microservices
  • FTP / file systems
  • Custom services

④ Integration layers

You can expose the entire AI agent as an API or an MCP server. Users and applications can then consume it in the AI agent interface of your choice, such as chat, SaaS apps, websites, or microservices.

AI agents exposed as an API can be embedded into applications, including App Builder applications, to create intelligent automation. Learn more in How to create an intelligent application.

⑤ AI accountability

Every interaction is wrapped in Jitterbit's AI accountability framework, which provides a complete audit trail for governance and security. This includes the user's prompt, the LLM's reasoning, and the tool's execution.