An open-source runtime where a persistent queen agent plans work, spawns parallel worker agents, and tracks results in a shared SQLite tracker.
agentscope-ai/agentscope-java
A Java framework for building long-running, distributed agents with event streaming, tool permissions, middleware and sandboxed execution.
- Stars
- 6,276
- Forks
- 1,575
- Open issues
- 972
- Last commit
- 11 hours ago
What is it?
AgentScope Java is a framework for building LLM-based agents in Java, targeting agents that run for a long time, serve many users and survive restarts. It provides a ReAct reasoning-and-acting core, plus a HarnessAgent layer that adds memory, skills, plan mode and subagent delegation on top of it.
The framework covers the runtime concerns that come up in production. Tool calls can be allowed, held for user approval or denied, and a unified event stream exposes model output, text deltas and tool execution in real time. Tools can run in local subprocesses, Docker, Kubernetes or an E2B cloud sandbox, and session state can be persisted to in-memory stores, JSON files, MySQL, Redis or PostgreSQL so that rolling deployments do not drop conversations.
Model providers ship as separate extension modules, including DashScope, OpenAI, Anthropic, Gemini and Ollama. Requires JDK 17 or higher and is distributed through Maven Central.
Key features
- 01Unified event stream with typed events for real-time rendering and human-in-the-loop.
- 02Tool-call permissions with allow, require approval, or deny states.
- 03Middleware hooks to extend the reasoning-acting loop.
- 04Tool execution in local, Docker, Kubernetes or E2B sandboxes.
- 05Session and memory persistence to Redis, MySQL, PostgreSQL and other stores for cross-replica recovery.
- 06Subagent spawning and messaging via agent_spawn and agent_send.
Use cases
- Backend services in Java that need agents handling multi-turn, long-running tasks.
- Enterprise applications requiring tool approval and audit of sensitive agent actions.
- Multi-tenant SaaS deployments that need isolated sessions and sandboxed code execution.
- Embedding agent workflows into existing Java systems with rolling deployments.
Quick start
- 1
Add the Maven dependency for agentscope-harness (JDK 17+)
- 2
Add a model provider extension, e.g. DashScope
- 3
Set DASHSCOPE_API_KEY and run the Hello AgentScope example
Things to know
- Requires JDK 17 or higher.
- Model providers are separate extension modules that must be added individually.
- Hello example needs a DashScope API key to run.
In collections
Topics
FAQ
- What is agentscope-java?
- A Java framework for building long-running, distributed agents with event streaming, tool permissions, middleware and sandboxed execution. AgentScope Java is a framework for building LLM-based agents in Java, targeting agents that run for a long time, serve many users and survive restarts. It provides a ReAct reasoning-and-acting core, plus a HarnessAgent layer that adds memory, skills, plan mode and subagent delegation on top of it.
- Is agentscope-java free to use? What license does it have?
- agentscope-java is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is agentscope-java still maintained?
- Actively maintained. The last commit to agentscope-java was on Nov 6, 2026.
- What can I use agentscope-java for?
- Backend services in Java that need agents handling multi-turn, long-running tasks. Enterprise applications requiring tool approval and audit of sensitive agent actions. Multi-tenant SaaS deployments that need isolated sessions and sandboxed code execution. Embedding agent workflows into existing Java systems with rolling deployments.
- How do I install agentscope-java?
- Add the Maven dependency for agentscope-harness (JDK 17+) Add a model provider extension, e.g. DashScope Set DASHSCOPE_API_KEY and run the Hello AgentScope example
- Where can I get agentscope-java?
- The source code and installation instructions are on GitHub: https://github.com/agentscope-ai/agentscope-java
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Code-first Python framework from Google for building, evaluating, and deploying AI agents, optimized for Gemini but model-agnostic.
Open-source long-horizon super agent harness that orchestrates sub-agents, memory, and sandboxes to research, code, and create across tasks lasting minutes to hours.
Self-hosted TypeScript agent runtime for coordinating multiple LLM agents, with human approval gates and offline-inspectable run journals.
Python framework for building AI agents that use LLMs and tools, and coordinate with each other in multi-agent workflows.
An open-source TypeScript framework and app stack for building autonomous AI agents, with a core runtime, plugins, and a local-first Eliza app.