AgentScope Java
https://github.com/agentscope-ai/agentscope-java- Category
- AI Agents
- Rank
- No. 501Tools index
- Pricing
- Open Source
- Type
- AGENT
- Builder
- agentscope-ai
- GitHub
- 5.5k stars
- Latest release
- v2.0.3
- Date
About
AgentScope's agent-oriented programming framework for the JVM — define agents, message passing, and orchestration in idiomatic Java.
What it does
It wraps a ReAct reasoning core with production machinery for long-running agents. Typed event streams expose each execution step, permission gates control tool use, and workspaces retain memory, skills, plans, and session state. Sandboxes and distributed state stores support isolated execution and recovery across processes.
Why it's ranked here
AgentScope Java is unusually complete for JVM teams moving beyond single-turn prototypes. Its strongest case is operational: event streaming, human approval, context compaction, tenant isolation, sandbox recovery, and distributed session persistence are framework concerns rather than application scaffolding. The trade is substantial platform scope and migration work from earlier APIs.
What's good
The core and harness are separable, so simple agents need not absorb workspace, persistence, or sandbox features. Model providers ship as optional modules. Typed events can drive live interfaces without manual response diffing. Middleware isolates logging, permissions, context injection, and business policy. State can persist through files, Redis, MySQL, or PostgreSQL.
Tradeoffs
It requires JDK 17 or newer. The second major generation introduces API-level breaking changes despite preserving compatibility where possible. Production deployments may involve external databases, object storage, containers, Kubernetes, or remote sandboxes. Provider integrations require separate dependencies, and the full harness carries more concepts than a basic reasoning loop.
How to use it well
Pick it for Java teams building long-running, multi-tenant agents that need recoverable sessions, controlled tools, live execution updates, or subagent delegation. Start with the core for a basic ReAct agent, then add the harness when persistence and isolation matter. The library does not replace the separate control plane and dashboard service.
Technical notes+
README.md presents agentscope-core as the minimal ReAct surface and agentscope-harness as the workspace, persistence, memory, sandbox, and orchestration layer. pom.xml defines a Maven multi-module build containing agentscope-core, agentscope-harness, agentscope-service, extensions, examples, BOMs, and distribution modules, compiled for Java 17 with JUnit, Mockito, Reactor Test, MockWebServer, Spotless, JaCoCo, and Surefire configuration. docs/v2/en/docs/index.md documents typed event streaming, RuntimeContext isolation keys, AgentStateStore backends, Markdown subagent specifications, context compaction, Plan Mode, and allow, approve, or deny tool permissions.
Observed
- License
- Apache License 2.0
- Primary language
- Java
- Packaging
- Multi-module Maven project with artifacts published through Maven Central
- Runtime requirement
- JDK 17 or newer
- Interface
- JVM library with reactive typed event streaming
- Protocols
- MCP and A2A support
- Model packaging
- OpenAI, Anthropic, Gemini, Ollama, and DashScope providers are separate extension modules
- Execution environments
- Local, Docker, Kubernetes, and remote cloud sandbox options are documented
Read from README.md, pom.xml, docs/README.md, docs/v1/en/intro.md, docs/v1/zh/intro.md, docs/v2/en/intro.md, docs/v2/zh/intro.md, docs/v2/en/docs/index.md, docs/v2/zh/docs/index.md, docs/v1/en/blogs/index.md, docs/v1/zh/blogs/index.md, docs/v2/en/blogs/usecases/index.md, docs/v2/zh/blogs/usecases/index.md, docs/v2/en/integration/rag/index.md, docs/v2/zh/integration/rag/index.md.
What it can do
Define autonomous agents
Agent specifications and behavior logic → Configured agent instances
Enable message passing between agents
Messages and target agent identifiers → Delivered messages to recipient agents
Orchestrate multi-agent workflows
Workflow definitions and agent coordination rules → Executed multi-agent processes
Create agent communication protocols
Communication patterns and message schemas → Established agent interaction frameworks
Build distributed agent systems
Agent deployment configurations and network topology → Running distributed agent applications
Implement agent lifecycle management
Agent startup, shutdown, and monitoring parameters → Managed agent states and processes
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.