Agent S开源图形用户界面代理框架，让大型语言模型（LLM）通过 Agent-Computer 接口像人类一样使用计算机。

4.6 (5)

审阅者 Daniel Nikulshyn·更新 2026年7月

Open Source AI Agents GUI Automation Framework LLM Desktop Automation Developer Tools

概览

Agent S 是一个开源的 GUI 代理框架，使大语言模型（LLMs）能够通过 Agent-Computer Interface 像人类一样与计算机交互。该框架让 LLMs 能够从过去的经验中学习，并在计算机上自主执行复杂任务。Agent S 设计用于单屏幕显示，支持 Linux、Mac 和 Windows 平台。该框架在多个基准测试上实现了最先进的表现，包括 OSWorld、WindowsAgentArena 和 AndroidWorld。最新版本 Agent S3 在 OSWorld 上的得分已突破人类水平，达到 72.60%。它同样展示了强大的 zero-shot 泛化能力。 Agent S 提供了灵活且模块化的架构，用于构建 GUI 代理。该框架包含一个名为 gui-agents 的库，用户可以轻松将 Agent S 与自己的应用集成。该库支持多平台，并提供简便的安装过程。 Agent S 的开发聚焦于提升自主 GUI 代理的能力。该框架有望用于包括自动化、AI 研究和计算机视觉在内的多种应用。不过，用户在运行 Agent S 时应保持谨慎，因为它通过运行 Python 代码来控制计算机。

主要功能

自主与计算机交互
Agent-Computer 接口
支持 Linux、Mac 和 Windows
基于 Python 的代码控制
行为最佳性能增强
GUI-agents 库

价格

模型: Free
分类: 网统粗子打得粗琞机
评分: 4.6 / 5 (5)

使用场景

自动化重复桌面工作流程

使用 LLM 驱动的代理在应用程序中导航 GUI、点击按钮和填写表单，消除日常计算机任务的手动重复。

构建自定义计算机使用代理

开发人员可以利用开源框架和 Agent-Computer 接口来原型和部署与操作系统交互的 LLM 代理，如同人类用户一样。

研究 GUI 代理能力

为学术界和 AI 研究人员提供了一个可复制的框架，用于基准测试和研究语言模型如何执行现实世界的计算机交互任务。

桌面应用的 QA 测试

部署一个代理来执行软件产品中的 GUI 工作流程，在不同场景中验证 UI 行为，而无需手动编写每个步骤的脚本。

优点 & 缺点

优点

在 OSWorld 上达到人类水平的表现
具有强大的零样本泛化能力
比以前的版本更简单、更快、更灵活

缺点

由于其能够控制计算机，需要谨慎使用
在多显示器屏幕上的使用限制尚不清楚

对决战绩

在万神殿中参与了 1 对决。

第1

第2

第3

Last battle

#4
AI Agent Development Frameworks Showdown — April 6, 2026
Apr 6, 2026 · #4 of 5

评测

4.6

5 个评分的平均值。

登录以留下评测。

Yuki Mori

Apr 28, 2026

Skeptical, then convinced

I went in skeptical — most tools in this space overpromise. It actually delivers on the API, and it is genuinely easy to set up caught me off guard. Pricing gets steep at scale is why this isn't a perfect score, still, I'd recommend giving it a real trial.

Leila Hassan

Apr 19, 2026

Compared a few options

Evaluated this against two competitors. Where it wins: the onboarding and it saves real time. Where it lags: a few rough edges remain. On balance the feature set — especially the dashboard — justifies the 5 stars for our use case.

Hiroshi Tanaka

Dec 1, 2025

Skeptical, then convinced

I went in skeptical — most tools in this space overpromise. It actually delivers on the onboarding, and the value for money is strong caught me off guard. The docs could be deeper is why this isn't a perfect score, still, I'd recommend giving it a real trial.

Olga Ivanova

Oct 4, 2025

Years in this space

I've evaluated a lot of these over the years. What stands out here is the dashboard — handled better than most — and the value for money is strong. Pricing gets steep at scale is my one real gripe. Worth the time if this is your use case.

Kwame Mensah

Jul 21, 2025

Years in this space

I've evaluated a lot of these over the years. What stands out here is the core workflow — handled better than most — and the value for money is strong. A few rough edges remain is my one real gripe. Worth the time if this is your use case.

问答

What are common use cases for Agent S?

Agent S is designed for tasks where an LLM needs to operate desktop GUI applications autonomously, such as automating workflows, interacting with software that lacks APIs, or research into computer-using AI agents via its Agent-Computer Interface.

Asked by Grace Okafor · Apr 25, 2026

Is Agent S free to use, and can I self-host or modify it?

Yes. Agent S is open-source, so you can use, self-host, and modify it according to its license terms. This makes it suitable for developers and researchers who want full control over the agent's behavior and integrations.

Asked by Kwame Mensah · Apr 21, 2026

What is Agent S and how does it interact with my computer?

Agent S is an open-source GUI agent framework that enables a large language model to operate your computer like a human user. It does this through an Agent-Computer Interface (ACI), allowing the LLM to perceive and control graphical applications.

Asked by Diego Fernández · Jan 31, 2026