AgentScope is a production-ready, open-source multi-agent framework designed for building transparent, understandable, and trustworthy AI agent systems. Developed by the agentscope-ai team, it provides essential abstractions for increasingly agentic LLMs with support for ReAct agent patterns, multi-agent orchestration via MsgHub, and human-in-the-loop steering with realtime interruption capabilities. The framework features realtime voice agent support for natural conversation interfaces, comprehensive memory management with both short-term and long-term memory systems plus memory compression, and full MCP (Model Context Protocol) integration for tool use. AgentScope supports Anthropic Agent Skills, A2A (Agent-to-Agent) protocol for inter-agent communication, and agentic reinforcement learning through Trinity-RFT integration. For observability, it leverages OpenTelemetry and provides SQLite-based session management. The platform is deployment-ready with Docker and Kubernetes support, includes browser automation capabilities, and integrates text-to-speech functionality. Built in Python 3.10+ with active biweekly community meetings and comprehensive documentation at doc.agentscope.io, AgentScope bridges the gap between research prototypes and production agent deployments.