又一个多 Agent 编程 IDE 神器,开源了!Another Multi-Agent Programming IDE Marvel Has Gone Open Source!


AI free token/AI免费积分

电脑上装了一堆 Agent 编程工具,有 Claude Code、Codex、Cursor 还都开了会员。

但在优化重构同个项目的时候,更多还是轮流让它们干活,怕出现代码冲突、覆盖文件等问题。

即便多开了几个 Agent 并行干活,来回切换查看进度、确认权限也是一件麻烦事。

效率不但没有提升上来,反倒给自己埋了不少坑,唯一好处就是几个 Agent 额度可以轮流用。

直到最近,在 GitHub 上闲逛时发现 Orca,一个专门为并行 Agent 做的开发环境。

image-20260709165620441

刚开源不久,已斩获 1.4 万 Star。官方叫它为 ADE,简单来说就是给 Agent 用的 IDE。

可以把一个需求派给多个 Agent 并行干活,每个 Agent 各改各的,不覆盖,不冲突。

这样做有个好处就是,同一个功能,分别让 Claude Code、Codex 或 Cursor 来开发。

当开发完成之后,我们可以对比它们各自实现的效果如何,哪个好就采用哪个。

image-20260709165849440

以前这样的做法,我们需要新建分支,切来切去进行对比,现在有 Orca 在一个面板解决。

除了 Claude Code、Codex 和 Cursor 外,还支持接入 GitHub Copilot、Qwen、Gemini 等等将近 30 个主流 Agent。

而且用的都是我们自己的订阅,不用额外花一分钱,项目也开源免费使用。

有个小细节,还会显示出 Claude Code、Codex 的用量和限流重置时间。

当额度快用完的时候,可以直接切换到另一个账号,无需重开会话,就连重新登录都不用。

image-20260709165955056

另一个我觉得挺实用的功能,是 Design Mode(设计模式)。

Web 应用修改页面样式的时候,直接在内置的 Chromium 窗口里点想改的那个元素。

这个元素位置的 HTML、CSS 代码,再加一张截图就自动传到对话框。

以前要跟 Agent 描述半天的修改,现在只需要点一下修改位置,再简单描述一下就行。

design-mode

至于代码审查,Agent 改完看到哪行代码有问题,直接在那行写条评论就能送回去让它改。

评论、改动、提交全部都可以在 Orca 里完成,不用再借助其他工具来回切换。

若有用 GitHub 或 Linear 工具,项目看板也能在 Orca 工具里直接查看。

还能直接对某个任务或者 issues,直接新建一个工作树就能直接开工,超级方便。

image-20260709170135133

甚至还做了个手机 App,在电脑上跑着的任务,人不在电脑前也能处理权限和下达指令。

若遇到非得在界面上点几下才能走下去的操作,还能让 Agent 接管桌面帮我们点。

剩下的实用功能还有很多,比如全局搜索、远程 SSH、终端分屏等等,就不一一介绍了。

image-20260709170218919

想上手体验的话,可以到项目的 GitHub 发布页面或官网上下载安装包。

支持 macOS、Windows 和 Linux 系统,另外界面也支持中文,对我们挺友好的。

若是喜欢用包管理器安装,macOS 系统可以用 brew,命令如下:

  •  
brew install --cask stablyai/orca/orca

每个功能在官网和 README 上里都配了直观的演示图,可以先去看下效果了解。

写在最后

现在大家都在用 AI 辅助写代码,但更多的还是只用一个 Agent,它在跑的时候便不敢动。

而有了 Orca,可以把从一个 Agent 变成管着一堆 Agent 同时干活。

比如同一个功能交给几个 Agent 并行去干,最终我们来对比选择一个实现最好的。

这个非常适合手上订阅了几个 Agent 的朋友,能让我们的干活效率拉满。

往后可能得看我们对各自 Agent 的能力了解程度,以及如何调度分配任务给它们干了。

毕竟模型再强也有跑偏的时候,多开几个 Agent 并行赛跑,总比单押一个要稳。

GitHub 项目地址:https://github.com/stablyai/orca

今天的分享到此结束,感谢大家抽空阅读,我们下期再见,Respect!

 

I have a ton of agent programming tools installed on my computer, including Claude Code, Codex and Cursor, all with premium subscriptions.

But when optimizing and refactoring the same project, I still tend to rotate them one at a time to avoid code conflicts and file overwrites.

Even if I run multiple agents in parallel, switching back and forth to check progress and confirm permissions is a hassle.

This doesn't improve efficiency; instead, it creates more problems, with the only benefit being that I can share my quota across agents.

Recently, I stumbled upon Orca while browsing GitHub — a development environment specifically for parallel agent orchestration.

image-20260709165620441

Newly open-source, it already has 14K stars. Officially called an ADE (Agent Development Environment), it's essentially an IDE for agents.

It lets me assign one task to multiple agents in parallel, each working on their own version without overwriting or conflicting.

This means I can have Claude Code, Codex or Cursor implement the same feature separately.

After completion, I compare their results and pick the best one — no more guesswork or trial-and-error.

image-20260709165849440

Previously, this required creating branches and switching between them for comparison; now Orca handles it all in one pane.

Besides the ones I mentioned, it supports GitHub Copilot, Qwen, Gemini and ~30 other mainstream agents.

Best of all, it uses my existing subscriptions — no extra cost. The project is also open-source and free to use.

A nice touch: it displays Claude Code and Codex usage limits and reset times.

When quota runs low, I can switch accounts without restarting the session or logging in again.

image-20260709165955056

Another useful feature is Design Mode, which simplifies web UI/UX changes.

In the built-in Chromium window, I click the element I want to edit.

Its HTML/CSS code and a screenshot are automatically sent to the chat window.

No more vague descriptions — just point and instruct.

design-mode

For code review, if an agent makes a mistake, I comment directly on that line and send it back for revision.

Comments, edits and commits are all handled within Orca — no need for external tools or context switching.

It even integrates GitHub and Linear boards, allowing me to create work trees from tickets or issues directly.

image-20260709170135133

image-20260709170135133

If an agent needs manual intervention (e.g., clicking a button), I can remotely control the desktop to do it.

Other features include global search, remote SSH, terminal splitting — but I won't cover them all.

image-20260709170218919

image-20260709170218919

It supports macOS, Windows and Linux, with a user-friendly Chinese interface.

On macOS, you can install via homebrew:

The website and README have plenty of demos to help you understand the features.

Final thoughts: while AI coding assistants are popular, most people only use one and pause when it's running.

With Orca, I can treat them like a team — running multiple agents in parallel and picking the best result.

This is ideal for anyone who subscribes to multiple agents.

The key now is understanding each agent's strengths and assigning tasks accordingly — no single-point failure risk.

GitHub: https://github.com/stablyai/orca

That's all for today. Thanks for reading! Until next time, respect!

Going forward, it will likely depend on how well we understand the capabilities of each respective Agent, and how we schedule and assign tasks to them.

After all, no matter how strong a model is, it can still go off track; running multiple Agents in parallel is steadier than betting on just one.

GitHub 项目地址:https://github.com/stablyai/orca

That’s all for today’s sharing, thanks for taking the time to read, see you next time, Respect!

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry