agentscope-ai/agentscope
Build and run agents you can see, understand and trust.
What it solves
AgentScope is a production-ready framework designed to build agentic applications that leverage the reasoning and tool-use capabilities of LLMs without restricting them with overly strict prompts or rigid orchestrations.
How it works
It provides a set of essential abstractions and building blocks to create flexible agents:
- Event System: A unified event bus that supports human-in-the-loop interactions and frontend integration.
- Permission System: Configurable, fine-grained control over how agents access tools and resources.
- Workspace/Sandbox: Isolated environments (supporting Docker, E2B, OpenSandbox, and Daytona) for running code and tools safely.
- Middleware System: Composable hooks that allow developers to customize the reasoning-acting loop of an agent.
- Agent Service: A FastAPI-based backend supporting multi-tenancy and multi-session isolation for production deployment.
Who it’s for
Developers building production-grade AI agents who need robust infrastructure for tool execution, session management, and human-in-the-loop control.
Highlights
- Production-Ready Serving: Built-in support for multi-tenancy and multi-session services.
- Longe-term Memory: Integration with ReMe and Mem0 for agentic memory.
- RAG Support: Built-in capabilities for Retrieval-Augmented Generation.
- Isolated Execution: Multiple sandbox backends for secure tool and code execution.
- Flexible Orchestration: Focuses on model reasoning over strict prompt constraints.