Skip to content

Minions: An Extensible Agentic AI Framework for Java

Minions is a powerful, MCP-compliant agentic AI framework designed for the enterprise Java ecosystem. Built with idiomatic Spring practices, it provides a robust foundation for creating and orchestrating autonomous AI agents in complex and regulated environments.

The framework is architected around a core philosophy:

The Open/Closed Principle: The system is open for extension but closed for modification.

This means you can build anything from a simple ReAct-style agent to a sophisticated planner-executor pipeline without ever needing to alter the core framework code.

Why Minions?

While many agentic frameworks exist for Python, Minions uniquely brings this cutting-edge capability to enterprise Java development. It's designed from the ground up to support the demands of modern, secure, and scalable applications.

  • Idiomatic Spring Integration: Leverages Spring Boot auto-configuration, dependency injection, and familiar idioms for seamless development.
  • Modular & Extensible: Easily add new components—tools, memory systems, or agent steps—as standard Spring Beans.
  • Governance by Design: Built with features like audit hooks, multi-tenant data isolation, and memory privacy for regulated industries.
  • Multi-Tiered Memory: Features a sophisticated memory architecture with pluggable strategies for querying, persistence, and summarization.
  • Enterprise-Grade Persistence: Integrates seamlessly with Spring Data for robust, scalable data storage.

Core Design Principles

  • Extensibility First Minions utilizes proven design patterns, such as the Factory Registry and Chain of Responsibility, to ensure seamless integration of custom components.

  • MCP Compliance Adherence to the Model Context Protocol (MCP) ensures architectural clarity, composability, and interoperability between different models and tools.

  • Enterprise Readiness With features like structured evaluation and memory management designed for privacy, Minions is ready for deployment in finance, healthcare, and insurance.

Sophisticated Memory Architecture

Minions features a pluggable, multi-tiered memory system that you can customize to your needs. Out of the box, it provides a powerful layered approach:

  • Short-Term Memory: Manages the active context for immediate tasks.
  • Episodic Memory: Stores a chronological history of events and interactions.
  • Vector Memory: Enables semantic search over vast amounts of information.
  • Long-Term Memory: Provides persistent, long-term knowledge storage.

Each subsystem is managed by pluggable strategies for querying, persistence, summarization, and privacy, giving you full control over how your agent remembers and forgets.

Agent Recipes

Every agent in Minions is configured using a Recipe. A recipe is a blueprint that defines the agent's behavior and capabilities.

A recipe specifies:

  • The agent's system prompt and ultimate goal.
  • The step graph, which outlines its reasoning and action flow.
  • Its complete memory configuration, including which subsystems to use and how they behave.
  • The toolchains and observers available to the agent.

Minions includes several pre-built recipes to get you started, such as ReActAgentRecipe and PlannerExecutorObserverRecipe, with a clear path to creating your own.