OpenDevin开源自主 AI 软件工程师为端到端编码任务

4.5 (6)

审阅者 Daniel Nikulshyn·更新 2026年7月

Open Source AI Agents Developer Tools Self-Hosted Code Generation Automation LLM

概览

OpenDevin 是一个由社区驱动的开源项目，旨在复制 Devin 这款自治 AI 软件工程师的能力。它提供了一个代理框架，能够编写、执行和调试代码，浏览网页，并与 shell 交互，以在最小的人类指导下完成复杂的开发任务。该平台面向想要实验代理工作流或将 AI 辅助集成到构建流程中的开发者而设计。它支持多种 LLM 后端，提供沙盒化运行时以安全执行代码，并配备基于 Web 的界面，用于监控代理动作并实时与代理协作。因为它是开源的，OpenDevin 可以自托管、定制和扩展，适合用于研究、内部工具以及致力于推进自主编码代理的贡献者。

主要功能

自主代码生成和执行
内置 shell 和浏览器工具
web UI 为代理互动
可交换 LLM 后端
沙箱化 Docker 运行时间
可扩展代理架构

价格

模型: Freemium
分类: AI Agents
评分: 4.5 / 5 (6)

使用场景

自主特性开发

将最终编码任务委托给 AI 代理，代理撰写、执行和调试代码在沙箱化环境中，并且仅在开发人员监督下进行。

试验机器人工作流

研究人员和开发人员可以使用可交换 LLM 后端和可扩展框架来模拟和延伸代理架构。

本地安装的 AI 编码辅助

关心数据隐私的团队可以在自己的基础设施上部署 OpenDevin 来获得 AI 软件工程帮助而不必将代码发送给第三方。

自动化调试和 Shell 任务

使用内置 shell 和浏览器工具，让代理来调查问题、运行命令和在整个项目中解除 bug 自动化。

优点 & 缺点

优点

完全开源和自托管
支持多个 LLM 提供商
沙箱化执行环境
活动社区和快速迭代
可自定义和扩展
支持多种 LLM 后端
自主代码生成和执行

缺点

需要技术设置和配置
性能依赖于选择的 LLM 和 API 成本
仍在实验阶段，有可靠性缺口
可能需要额外的成本
依赖高级技能进行操作

评测

4.5

6 个评分的平均值。

登录以留下评测。

Joanna Kowalski

May 2, 2026

Solid for our team

We rolled this out across the team last quarter and sandboxed execution environment. Sandboxed Docker runtime fits neatly into how we already work, and autonomous code generation and execution removed a step we used to do by hand. Still experimental with reliability gaps, which is the main caveat, but it has held up under daily use.

George Papadakis

Feb 28, 2026

Solid for our team

We rolled this out across the team last quarter and active community and rapid iteration. Pluggable LLM backends fits neatly into how we already work, and autonomous code generation and execution removed a step we used to do by hand. but it has held up under daily use.

Tomáš Novák

Nov 23, 2025

Does the job

Pretty happy overall. Pluggable LLM backends just works and fully open source and self-hostable. but no dealbreakers — I'd recommend it to a friend without hesitating.

Nadia Petrova

Nov 15, 2025

Does the job

Pretty happy overall. Sandboxed Docker runtime just works and active community and rapid iteration. Still experimental with reliability gaps can be annoying, but no dealbreakers — I'd recommend it to a friend without hesitating.

Carlos Mendoza

Nov 12, 2025

Years in this space

I've evaluated a lot of these over the years. What stands out here is autonomous code generation and execution — handled better than most — and supports multiple LLM providers. Requires technical setup and configuration is my one real gripe. Worth the time if this is your use case.

Aaliyah Johnson

Sep 17, 2025

Solid for our team

We rolled this out across the team last quarter and supports multiple LLM providers. Web UI for agent interaction fits neatly into how we already work, and autonomous code generation and execution removed a step we used to do by hand. Still experimental with reliability gaps, which is the main caveat, but it has held up under daily use.