agentscope-java

Agentscope‑Java lets you create and run AI‑driven agents directly from Java or Spring Boot applications. You can either embed the Use SDK in your code for full control, or call a managed Agent service that runs the runtime for you. The site positions the tool for teams that need production‑grade, long‑running agents that can be integrated into existing Java services.

The Use SDK provides a set of building blocks: context handling, workspaces, tool integration, persistent sessions, and a reasoning loop. You compose these pieces in code, configure a builder with a name, model, and workspace path, then create an agent instance per request. The example shows creating a `HarnessAgent.builder()` with a model identifier such as `dashscope:qwen-plus`, adding a `RuntimeContext` that includes a user and session ID, and streaming events to process a user message. The SDK also supports sandboxes, sub‑agents, event recording, and checkpoint recovery, which are useful for complex workflows that require stateful interactions.

If you prefer not to manage the runtime, the AgentScope Service offers an HTTP API. You configure an agent definition, model, tools, and resources, through the platform, then submit tasks or conversational turns via `/api/v1/agent-sessions` endpoints. The service handles execution, session persistence, and result delivery, allowing your application to focus on business logic and data handling. The documentation mentions that managed agents are built on the same Use core, so the behavior is consistent between the two approaches. Both options can be combined: you might embed a custom agent for specialized processing while also calling managed agents for generic tasks.

The platform claims production use at Alibaba Group across many business units and open‑source adoption in multiple sectors, indicating that the framework has been exercised in large‑scale environments. The site provides quick‑start code snippets for both the SDK and the service, as well as links to source on GitHub and a documentation index file (`/llms.txt`) for exploring available pages.

Key features

• Use SDK for deep customization of agent behavior in Java code
• Managed Agent service with HTTP API for runtime‑free integration
• Persistent sessions, checkpoints, and event streaming for long‑running tasks
• Context and workspace management to load memory, skills, and files on demand
• Tool and sandbox integration, including sub‑agent delegation
• Shared execution core between SDK and managed service for consistent behavior