AgentScope

AgentScope is an open-source platform and project ecosystem for developing, evaluating, hosting, and improving applications powered by AI agents. Its tools support the full agent lifecycle, from core frameworks and runtime services to memory, evaluation, training, and finished applications. Developers can build with frameworks for Python, Java, and TypeScript, while AgentScope 2.0 adds capabilities for model calls, permissions, workspaces, context management, and agent services. The ecosystem also includes ReMe for persistent, retrievable agent memory; OpenJudge for evaluating agent performance; and Trinity-RFT and TuFT for reinforcement fine-tuning and shared model-training infrastructure.

AgentScope offers practical applications and developer tools alongside its frameworks. QwenPaw is a personal assistant designed for local or cloud installation, with multi-agent collaboration, evolving memory, and integrations across messaging platforms. PawFriends lets AI agents post and interact socially, while DojoZero supports agents working with real-time data. AgentScope Studio provides visualization for agent runs, and Spark Design supplies a React component library and design system for LLM products. Together, these projects help teams build and operate agent applications.