# AgentScope review

> AgentScope is a Python agent framework with an SDK for building ReAct agents and an agent service backend with a web UI.

Devtoolwatch Board, the week of Oct 25–31, 2026. Canonical page: https://index.devtoolwatch.com/tools/agentscope-ai-agentscope/

## TL;DR

AgentScope 2.0 is an open-source Python framework for building, running and serving LLM agents. It is for developers who want both an agent SDK and a multi-tenant, multi-session backend with a pre-built Web UI. It adds permission controls, human-in-the-loop confirmation and sandboxed tool execution.

## AgentScope at a glance

- **Category:** [Agent frameworks](https://index.devtoolwatch.com/categories/agent-frameworks/)
- **Maintained by:** AgentScope-AI (organization, [@agentscope-ai](https://github.com/agentscope-ai) on GitHub)
- **License:** Apache 2.0
- **Language:** Python, Dockerfile, Mako
- **Repository:** [github.com/agentscope-ai/agentscope](https://github.com/agentscope-ai/agentscope)
- **Website:** [docs.agentscope.io](https://docs.agentscope.io/)
- **PyPI package:** [agentscope](https://pypi.org/project/agentscope/) 2.0.10
- **Latest release:** [v2.0.10](https://github.com/agentscope-ai/agentscope/releases/tag/v2.0.10) on GitHub, Oct 12, 2026
- **First published:** Jan 12, 2024

## What AgentScope does

- AgentScope runs ReAct agents with structured output, realtime interruption and resume, and sequential or concurrent tool calls.
- AgentScope manages Python tools, MCP servers and skills, and ships built-in shell, file edit, search, task and plan tools.
- AgentScope supports models from OpenAI, Anthropic, Gemini, DashScope, DeepSeek, Moonshot, Volcengine, xAI and Ollama.
- AgentScope runs tools and code in isolated workspaces: local, Docker, Apple Container, Bubblewrap, E2B, OpenSandbox, Daytona or K8s.
- AgentScope supports multi-agent setups through TeamPipeline, where a leader agent delegates tasks to member agents that run concurrently.
- AgentScope connects to remote agents over the A2A protocol through the A2AAgent class.
- AgentScope offers experimental realtime voice agents using DashScope, OpenAI, Gemini and xAI realtime APIs.

### How it works

AgentScope is a Python package. You compose an agent from a model, a toolkit and middleware, then chat with it in the terminal using launch_console. For applications, the agent service provides a FastAPI backend and a pre-built Web UI. The service adds multi-tenancy, SQL or NoSQL persistence, scheduling, RAG and IM channels such as Feishu and Discord.

## How to install AgentScope

**From PyPI**

```
uv pip install agentscope
```

**From source**

```
git clone -b main https://github.com/agentscope-ai/agentscope.git
```

**Editable install from source**

```
uv pip install -e .
```

### Requirements

- Python 3.11 or higher
- An API key for a supported model provider, such as DASHSCOPE_API_KEY in the quickstart
- pnpm to run the web UI

### First steps

1. Install AgentScope on Python 3.11 or newer with uv pip install agentscope.
2. Set the DASHSCOPE_API_KEY environment variable for the quickstart's DashScope model.
3. Create an Agent with a model and a Toolkit, then call launch_console to chat in the terminal.
4. To run the agent service, start examples/agent_service with python main.py. Then run pnpm install and pnpm dev in examples/web_ui.

## Who AgentScope is for

- Python developers building tool-using LLM agents with permission controls and sandboxed execution
- Teams shipping multi-tenant, multi-session agent apps with a backend and a web UI
- Projects that need multi-agent orchestration with leader and worker agents
- Developers connecting agents to Feishu or Discord, or building realtime voice agents

### Limitations

- AgentScope 2.0 is a breaking change from 1.0, so existing 1.0 code needs migration.
- SOP, realtime voice agents and WebRTC voice in the agent service are marked experimental.
- AgentScope requires Python 3.11 or higher.

## Usage data (the week of Oct 25–31, 2026)

- Board rank: #45 overall, #3 in Agent frameworks, momentum score 77
- Downloads that week: 51.4k (+38% vs 4-week average)
- New GitHub stars per week (last 12 weeks): Aug 9: 256; Aug 16: 333; Aug 23: 773; Aug 30: 742; Sep 6: 695; Sep 13: 509; Sep 20: 449; Sep 27: 317; Oct 4: 354; Oct 11: 402; Oct 18: 460; Oct 25: 1.2k

## AgentScope alternatives

| Rank | Tool | Stars | Downloads/wk |
|---|---|---|---|
| #1 | [Harness²](https://index.devtoolwatch.com/tools/google-research-harness2/) | 21.7k | – |
| #2 | [HyperFrames](https://index.devtoolwatch.com/tools/heygen-com-hyperframes/) | 76.6k | 1.5M |
| #4 | [AG-UI](https://index.devtoolwatch.com/tools/ag-ui/) | 16.9k | 4.6M |
| #5 | [LangGraph](https://index.devtoolwatch.com/tools/langgraph/) | 44.2k | 18.5M |
| #6 | [Bifrost](https://index.devtoolwatch.com/tools/maximhq-bifrost/) | 9.3k | – |
| #7 | [assistant-ui](https://index.devtoolwatch.com/tools/assistant-ui/) | 12.9k | 2.8M |

## AgentScope FAQ

### Is AgentScope free and open source?

Yes. AgentScope's source code is on GitHub (agentscope-ai/agentscope) under the Apache 2.0 license, so you can use it for free.

### How do I install AgentScope?

Run `uv pip install agentscope`. The agentscope package needs Python 3.11 or newer. The install section above has the other options from the project's README.

### Is AgentScope still maintained?

Yes. The agentscope-ai/agentscope repository was last updated on Oct 21, 2026, and the latest release, v2.0.10 on GitHub, came out on Oct 12, 2026.

### How popular is AgentScope?

AgentScope has 35k GitHub stars as of Nov 1, 2026. It gained 1.2k new stars in the week of Oct 25–31, 2026, against a 12-week average of 541 a week. Its main package had 51.4k downloads that week, +38% against its own 4-week average. That made it #3 among AI agent frameworks on the Devtoolwatch Board.

### AgentScope vs LangGraph: how do they compare?

AgentScope and LangGraph are both tracked by Devtoolwatch as AI agent frameworks. AgentScope has 35k GitHub stars and LangGraph has 44.2k. In the week of Oct 25–31, 2026, AgentScope's main package had 51.4k downloads and LangGraph's had 18.5M. LangGraph is a low-level orchestration framework and runtime for building, managing, and deploying long-running, stateful agents in Python or JavaScript. [LangGraph review](https://index.devtoolwatch.com/tools/langgraph/)

### AgentScope vs LangChain: how do they compare?

AgentScope and LangChain are both tracked by Devtoolwatch as AI agent frameworks. AgentScope has 35k GitHub stars and LangChain has 148.8k. In the week of Oct 25–31, 2026, AgentScope's main package had 51.4k downloads and LangChain's had 41.1M. [LangChain review](https://index.devtoolwatch.com/tools/langchain/)

### What are the best alternatives to AgentScope?

Other AI agent frameworks on the Devtoolwatch Board in the week of Oct 25–31, 2026, by momentum: Harness² (#1), HyperFrames (#2), AG-UI (#4), LangGraph (#5), Bifrost (#6). The alternatives table above has their numbers.

### What is AgentScope 2.0?

AgentScope 2.0 is a Python agent framework with two parts. The SDK covers agents, tools, models, context, memory and sandboxes. The agent service is a FastAPI backend with a Web UI that supports multi-tenant, multi-session apps and distributed deployment.

### Which model providers does AgentScope support?

AgentScope supports LLM, embedding and TTS models from OpenAI, Anthropic, Gemini, DashScope, DeepSeek, Moonshot, Volcengine, xAI and Ollama. It also supports MiniMax chat models through MiniMaxChatModel. ModelRouterMiddleware can choose a model for each reply.

### Does AgentScope support MCP?

AgentScope's Toolkit manages MCP servers alongside Python tools and skills. The agent service also includes an MCP & Skill Hub for browsing the GitHub MCP Registry and ClawHub and installing servers into a workspace.

### How does AgentScope keep agent tool use safe?

AgentScope describes three layers of protection: tool-specific review, human-in-the-loop permissions and sandboxing. You can require confirmation for tool calls or use a bypass mode. Tools run in isolated workspaces, separated at the user, agent or session level.

### Can AgentScope run multi-agent systems?

AgentScope supports multi-agent systems. TeamPipeline lets a leader agent delegate tasks to member agents that run concurrently. The agent service adds leader-worker orchestration with built-in team tools. A2AAgent lets agents talk to remote A2A agents.

## Sources

1. [agentscope-ai/agentscope on GitHub](https://github.com/agentscope-ai/agentscope)
2. [AgentScope README](https://github.com/agentscope-ai/agentscope/blob/main/README.md)
3. [agentscope on PyPI](https://pypi.org/project/agentscope/)
4. [AgentScope website](https://docs.agentscope.io/)
5. [Release v2.0.10 (Oct 12, 2026)](https://github.com/agentscope-ai/agentscope/releases/tag/v2.0.10)
6. [Release v2.0.9 (Sep 28, 2026)](https://github.com/agentscope-ai/agentscope/releases/tag/v2.0.9)
7. [Release v2.0.8 (Sep 8, 2026)](https://github.com/agentscope-ai/agentscope/releases/tag/v2.0.8)
8. [Devtoolwatch Board, the week of Oct 25–31, 2026](https://index.devtoolwatch.com/)