agentscope
Build and run agents you can see, understand and trust.
What is agentscope?
Build and run agents you can see, understand and trust.
Overview of agentscope
What it is
AgentScope 2.0 is an open-source agent framework that its README describes as production-ready and easy to use, built on abstractions intended to keep pace with rising model capability. The project states that it is designed for increasingly agentic LLMs, leveraging models' reasoning and tool-use abilities rather than constraining them with strict prompts and opinionated orchestrations. It requires Python 3.11 or higher. The repository ships both an SDK layer for composing agents and a batteries-included agent service.
Key features
- ReAct reasoning-acting loop with structured output, realtime interruption and resume, and batched sequential or concurrent tool acting.
- Toolkit for agentic tool management over Python tools, MCP servers and skills, with built-in coding tools (shell, file edit, search) and task/plan tools.
- Model support across providers including OpenAI, Anthropic, Gemini, DashScope, DeepSeek, Moonshot, Volcengine, xAI and Ollama, plus embedding and TTS.
- Context handling with automatic compaction, tool-result offload and context injection (system prompt, RAG, memory) through built-in middleware.
- Unified event bus streaming reasoning, tool calls and multimodal content (text, image, audio) to the frontend.
- Permission and human-in-the-loop control, composable middleware hooks, and agentic memory with switchable backends (ReMe, Mem0).
- Workspace and sandbox isolation for tool and code execution: local, Docker, Apple Container, Bubblewrap, E2B, OpenSandbox, Daytona and K8s.
Use cases
The framework targets agents that reason and call tools, including coding tools run inside isolated workspaces. The agent service is a FastAPI backend with a pre-built Web UI that turns agents into a multi-tenant, multi-session application, with leader-worker agent teams, channels for Feishu (Lark), Discord and custom platforms, a RAG service, an MCP and Skill Hub, resource sharing, SQL and NoSQL persistence, and scheduling with background task offloading. The README also lists experimental realtime voice agents over WebRTC and A2A protocol support for chatting with remote A2A agents.
Getting started
bash uv pip install agentscope From source: bash git clone -b main https://github.com/agentscope-ai/agentscope.git cd agentscope uv pip install -e . The README then shows an example that builds an Agent with a DashScopeChatModel and a Toolkit of built-in tools, and chats with it in the terminal via launch_console(agent). To run the agent service backend and web UI, the README provides commands under examples/agent_service and examples/web_ui.
License and pricing
AgentScope is released under the Apache License 2.0. It is open source; no pricing is stated.
Based on the project's public README and repository metadata.