agentscope-ai/agentscope-java

What it solves

AgentScope Java 2.0 is designed for building distributed, enterprise-grade AI agents that can handle long-running, complex tasks. It moves beyond simple reasoning loops to provide the infrastructure needed for production deployment, including session recovery, secure tool execution, and multi-tenant isolation.

How it works

The framework uses a dual-layer architecture. A core reasoning engine (like ReActAgent) is wrapped by a "Harness" that adds production capabilities via middleware and toolkit channels. It employs a unified event stream for real-time rendering and human-in-the-loop (HITL) interactions, and a permission system to gate tool calls. For distributed scaling, it uses stateless agents with external state stores (Redis, MySQL, PostgreSQL) for session recovery.

Who it’s for

Java developers building AI agents for enterprise environments who require high reliability, distributed deployment, and strict control over agent actions.

Highlights

• Harness Engineering: Includes self-evolving skill repositories, layered memory with auto-compaction, and the ability to spawn and delegate to sub-agents.
• Enterprise Deployment: Supports multi-tenant isolation and secure sandboxing via Docker, Kubernetes, or cloud sandboxes.
• Distributed State: Enables zero-downtime rolling deploys through cross-replica session recovery using various database backends.
• Event-Driven Architecture: A system of 31 typed events for real-time UI updates and a first-class Human-in-the-Loop system for approving sensitive actions.
• AgentScope Service: A separate control plane for agent registration, discovery, and orchestration across different runtimes like LangChain and ADK.