✨ Key Features
- Deployment Infrastructure: Built-in services for session management, memory, and sandbox environment control
- Sandboxed Tool Execution: Isolated sandboxes ensure safe tool execution without system compromise
- Developer Friendly: Simple deployment with powerful customization options
- Framework Agnostic: Not tied to any specific framework. Works seamlessly with popular open-source agent frameworks and custom implementations
- 🚧 Observability: Trace and visualize agent operations comprehensively (under development)
💬 Community
Join our community on DingTalk:
| DingTalk |
 |
📋 Table of Contents
🚀 Quick Start
Prerequisites
- Java 17 or higher
- Maven 3.6+
Add Dependency
Add the following dependency to your pom.xml:
<dependency>
<groupId>io.agentscope</groupId>
<artifactId>spring-boot-starter-runtime-a2a</artifactId>
<version>1.0.2</version>
</dependency>
<!-- Add AgentScope Agent adapter dependency -->
<dependency>
<groupId>io.agentscope</groupId>
<artifactId>agentscope-runtime-agentscope</artifactId>
<version>1.0.2</version>
</dependency>
Basic Agent Usage Example
The following example demonstrates how to delegate a AgentScope ReactAgent using AgentScope Runtime. The complete source code can be found in the examples directory.
- Create Agent Handler
Create a custom agent handler by extending AgentScopeAgentHandler:
public class MyAgentScopeAgentHandler extends AgentScopeAgentHandler {
@Override
public Flux<io.agentscope.core.agent.Event> streamQuery(AgentRequest request, Object messages) {
// Create Toolkit and register tools
Toolkit toolkit = new Toolkit();
if (sandboxService != null) {
Sandbox sandbox = sandboxService.connect(
request.getUserId(),
request.getSessionId(),
BaseSandbox.class
);
toolkit.registerTool(ToolkitInit.RunPythonCodeTool(sandbox));
}
// Create ReActAgent with tools
ReActAgent agent = ReActAgent.builder()
.name("Friday")
.toolkit(toolkit)
.model(DashScopeChatModel.builder()
.apiKey(System.getenv("AI_DASHSCOPE_API_KEY"))
.modelName("qwen-max")
.stream(true)
.formatter(new DashScopeChatFormatter())
.build())
.build();
// Convert messages and stream agent responses
// See examples/README.md for complete implementation
return agent.stream(queryMessage, streamOptions);
}
}
- Initialize Services and Deploy
Configure the agent handler with required services and deploy using AgentApp:
// Create and configure the agent handler
MyAgentScopeAgentHandler agentHandler = new MyAgentScopeAgentHandler();
agentHandler.setStateService(new InMemoryStateService());
agentHandler.setSessionHistoryService(new InMemorySessionHistoryService());
agentHandler.setMemoryService(new InMemoryMemoryService());
agentHandler.setSandboxService(new SandboxService(
new SandboxManager(ManagerConfig.builder().build())
));
// Deploy using AgentApp
AgentApp agentApp = new AgentApp(agentHandler);
agentApp.run(8090); // Server will listen on port 8090
Note
You can also use Kubernetes or Alibaba FC platform AgentRun to execute sandbox tools. Please refer to the examples directory for more details.
🔌 Agent Framework Integration
AgentScope Runtime Java implementation can be easily integrated with any agent frameworks developed in Java. Currently supported frameworks include:
- AgentScope Java
- Spring AI Alibaba, Langchain4j and more coming soon...
🤝 Contributing
We welcome contributions from the community! Here's how you can help:
🐛 Bug Reports
- Use GitHub Issues to report bugs
- Include detailed reproduction steps
- Provide system information and relevant logs
💡 Feature Requests
- Discuss new ideas in GitHub Discussions
- Follow the feature request template
- Consider implementation feasibility
🔧 Code Contributions
- Fork the repository
- Create a feature branch (
git checkout -b feature/amazing-feature)
- Commit your changes (
git commit -m 'Add amazing feature')
- Push to the branch (
git push origin feature/amazing-feature)
- Open a Pull Request
For detailed contributing guidelines, please see CONTRIBUTING.md.
📄 License
This project is licensed under the Apache License 2.0. See the LICENSE file for details.