agentscope-java is a Java framework for building and running distributed, production-grade, long-running agents in enterprise environments. It provides abstractions for safely controlled execution, multi-tenant isolation, and distributed horizontal scaling.
Project overview
It offers first-class human-in-the-loop and permission controls for safely governing long-running agent executions, addressing the gap between prototyping and deploying controlled production agents.
Project type
AI Agent
Deployment
Refer to project documentation
License
License pending
Best for
Enterprise teams building distributed, long-running agents that require multi-tenant isolation and safely controlled tool execution
Developers who need fine-grained permission gating and human-in-the-loop approval for agent tool calls
Key capabilities
A tool-call gating engine that can allow, require user approval, or deny tool executions.
AOP-style hook interception for flexibly extending the reasoning-acting loop across five stages.
Multiple subagent definition patterns with agent_spawn / agent_send and real-time event forwarding.
Infrastructure layers built on ReActAgent including self-evolution, layered memory, auto context management, and plan mode for long tasks.
Limitations and risks
The framework requires JDK 17 or higher to run, which may constrain environments with older Java runtimes.
Getting started
Ensure JDK 17 or higher is installed, then add the framework as a dependency via Maven. You also need to pick and add specific model provider extensions, such as Gemini models via an optional extension.