agentscope-ai/agentscope

AgentScope is a Python framework for building, running, and deploying AI agents, software programs powered by large language models (LLMs) that can reason, use tools, remember past interactions, and complete multi-step tasks autonomously. Its stated goal is to make agents you can "see, understand and trust." The framework is designed to adapt to increasingly capable AI models by leaning on their built-in reasoning and tool-use abilities rather than forcing them into rigid, pre-scripted workflows. It comes with built-in components that handle the common pieces of agent development: a ReAct agent (a standard pattern where the model reasons then acts, alternating between thinking and tool calls), memory (including long-term memory with database support), planning, tool execution, human-in-the-loop interaction (where a person can intervene or guide the agent mid-task), and real-time voice interaction. For multi-agent systems, where multiple AI agents collaborate or compete, AgentScope provides a message hub for flexible orchestration. It supports MCP (a standard protocol for connecting AI tools) and A2A (Agent-to-Agent, a protocol for agents to communicate with each other). You can also fine-tune the underlying models using reinforcement learning. Developers building chatbots, autonomous assistants, data processing pipelines, or complex multi-agent workflows use AgentScope. It can be deployed locally, as a serverless cloud function, or on a Kubernetes cluster. It requires Python 3.10 or higher and is licensed under Apache 2.0.