又一个并行 Agent ,终于有人做好了。Another parallel agent, finally someone did it right.


AI free token/AI免费积分

本文来源:https://mp.weixin.qq.com/s/bOi3qFv50RpAa-DnBt4Puw
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

 

很多开发者都在用 Claude Code、Codex 这些 Agent,都会面临这样的问题。

Agent 跑起来之后想要切换到另一个 Agent 去试一试,但是终端窗口切换起来很麻烦。

想同时运行多个Agent来比较方案,但是一个终端只能运行一个。

跑了跑了就出去了,回来的时候不知道Agent有没有跑完。

最近刷到了一个非常实用的工具。

Orca 是一个多Agent并行编排工具。

它在 GitHub 上已经有 2 万多人关注了。

图片

01 开源项目简介

Orca 可以让你在一个界面上同时运行多个 Agent,例如 Claude Code、Codex、Grok、Pi 等等 CLI Agent。

图片

每个 Agent 都在自己独立的工作树中运行,彼此之间不会互相影响,不会互相覆盖文件,也不会污染主分支。

可以将一个需求同时发送给三个Agent,让它们各自写出一个版本的代码,再进行比较,选择最好的一个。

执行过程中,手机端可以收到进度通知,出门也不需要一直盯着电脑看,可以随时掌握执行情况。

Orca 还可以和 GitHub、Linear 等项目管理工具进行原生集成,在工具中直接打开 PR、Issue,从任务创建 worktree。

02 我先用真实案例演示一把

比如我有一个电商后台管理系统的订单处理模块需要重构。

项目里有订单列表页面、支付逻辑、库存同步代码,文件都在仓库里,代码量不小。

我想重新构建这个模块来提高性能和错误处理,但是不知道从哪里开始,而且代码修改涉及到很多文件。

用上 Orca。

在命令行中输入:帮我把订单处理模块重构一下,提高性能和错误处理能力。

图片

然后Orca自动创建了3个并行的worktree,将同一个prompt分发给Claude Code、Codex、Pi这三个Agent。

每个Agent独立开工,各自实现重构方案,一个优化性能,一个加强错误处理,一个兼顾两者,互不干扰。

执行的时候,我去外面办了一些事,手机上收到了 Orca 发来的消息,说有一个 Agent 运行完毕了。

图片

打开手机端的Orca Companion,可以看到三个Agent的执行进度,并且可以远程发送后续指令。

图片

三个Agent都跑完了,Orca会自动展示出三个方案的差异。

经过比较之后发现,Pi 版本的性能优化更好一些,Claude Code 的错误处理更周到一些,Codex 的代码风格也更简洁一些,各有千秋。

我直接在Orca中给Claude Code的版本加上了一些注释,并且让其补充上性能优化的内容。

最后选择了 Pi 的版本,合并到主分支上,git commit 一步到位。

整个过程你只说了一句,其余的worktree创建、agent分发、diff对比、merge操作都是Orca自己完成的。

03 几个核心功能

① 并行Worktree

一个prompt同时分发给多个Agent,每个Agent各自在自己的worktree中运行,互不影响。

执行完自动对比diff之后选择最好的merge,避免了手动打开多个窗口、逐行比较的时间。

② 手机端 Companion

在Agent执行的过程中,手机端可以接收到进度通知,并且还可以远程发送后续指令。

图片

出门的时候不需要一直盯着电脑看,回来之后就知道有没有跑完。

③ 多Agent兼容

支持Claude Code、Codex、Grok、Cursor、GitHub Copilot、OpenCode、Pi 等 20 多个 CLI Agent,包括大部分主流的Agent。

图片

可以接入任意的Agent,不需要更换工具,也不需要学习新的使用方法。

④ AI Diff 批注

在Agent生成的diff中添加注释,并提出修改意见,不需要切换到另一个工具里去写批注。

图片

Review、Edit、Commit 一口气完成,从审核到提交都不需要离开工具。

⑤ 设计模式

点击元素就会自动生成完整的上下文,AI 会直接理解到 UI 的结构。

图片

⑥ 安装方便,使用简单

Orca 支持 macOS、Windows、Linux 三个桌面平台以及 iOS、Android 手机端。

图片

MacOS 用户可以使用 Homebrew 直接安装:

brew install --cask stablyai/orca/orca

安装好之后打开Orca,连接上你常用的Agent,就可以开始使用了,学习成本很低。

04 写在最后

说白了,它就是帮开发者解决多 Agent 管理的问题。

以前同时运行多个Agent的时候,要打开很多个终端窗口,来回切换很麻烦。

Orca 把它们都放在一个界面里,每个 Agent 都有自己的 worktree,互不干扰。

但是目前还存在一些限制,只支持CLI Agent,图形界面的Agent还不能接入。

手机端的Companion功能也在不断完善中,但是用来查看进度、远程下发任务已经足够了。

整体体验还是不错的。

感兴趣的话,可以装一个试试看。

开源地址:https://github.com/stablyai/orca

 

 

Article source: https://mp.weixin.qq.com/s/bOi3qFv50RpAa-DnBt4Puw If there is a copyright issue, please notify us by email: 13951431913#139.com or submit a comment on the right.

Many developers are using agents like Claude Code and Codex, and they all face this problem.

Once an agent starts running, if you want to switch to another agent to try it out, switching terminal windows is cumbersome.

You want to run multiple agents simultaneously to compare approaches, but one terminal can only run one.

You step out while it's running, and when you come back, you don't know whether the agent has finished.

Recently I came across a very useful tool.

Orca is a multi-agent parallel orchestration tool.

It has over 20,000 stars on GitHub.

01 Open Source Project Overview

Orca lets you run multiple agents simultaneously in one interface, such as Claude Code, Codex, Grok, Pi, and other CLI agents.

Each agent runs in its own independent worktree, so they don't affect each other, overwrite each other's files, or pollute the main branch.

You can send one requirement to three agents at the same time, have each produce a version of the code, then compare them and pick the best one.

During execution, you can receive progress notifications on your mobile phone, so you don't need to keep staring at your computer when you're out; you can keep track of execution anytime.

Orca also natively integrates with project management tools like GitHub and Linear, letting you open PRs and Issues directly in the tool and create worktrees from tasks.

02 Let me demonstrate with a real-world example.

For example, I have an order processing module in an e-commerce backend management system that needs refactoring.

The project has an order list page, payment logic, and inventory synchronization code; the files are all in the repository, and the codebase is not small.

I want to rebuild this module to improve performance and error handling, but I don't know where to start, and the changes touch many files.

That's where Orca comes in.

In the command line, I type: "Refactor the order processing module for me to improve performance and error handling."

Orca then automatically created 3 parallel worktrees and distributed the same prompt to three agents: Claude Code, Codex, and Pi.

Each agent started working independently, each implementing its own refactoring approach: one optimized performance, one strengthened error handling, and one balanced both—without interfering with each other.

While it was running, I went out to take care of some things, and I received a message from Orca on my phone saying that one agent had finished.

Opening Orca Companion on mobile shows the execution progress of all three agents, and you can send follow-up commands remotely.

When all three agents finish, Orca automatically shows the differences among the three approaches.

After comparing them, I found that the Pi version had better performance optimization, Claude Code's error handling was more thorough, and Codex's code style was cleaner—each has its strengths.

I directly added some comments to the Claude Code version in Orca and asked it to include performance optimizations.

In the end, I chose the Pi version, merged it into the main branch, and committed with git in one step.

Throughout the entire process, you only said one sentence; the rest—worktree creation, agent distribution, diff comparison, and merge operations—were all done by Orca itself.

03 Key Features

① Parallel Worktrees

One prompt is distributed to multiple agents at the same time, and each agent runs in its own worktree without affecting others.

After execution, it automatically compares diffs and picks the best one to merge, saving you from manually opening multiple windows and comparing line by line.

② Mobile Companion

While agents are executing, you can receive progress notifications on mobile and send follow-up commands remotely.

When you go out, you don't need to keep watching your computer; when you come back, you'll know whether it has finished.

③ Multi-Agent Compatibility

It supports 20+ CLI agents, including Claude Code, Codex, Grok, Cursor, GitHub Copilot, OpenCode, Pi, and most other mainstream agents.

You can connect any agent without changing tools or learning new usage methods.

④ AI Diff Annotations

Add comments to diffs generated by agents and propose changes without switching to another tool to write annotations.

Review, edit, and commit are done in one go—no need to leave the tool from review to commit.

⑤ Design Patterns

Click an element and it automatically generates full context, allowing AI to directly understand the UI structure.

⑥ Easy Installation and Simple Usage

Orca supports macOS, Windows, and Linux desktop platforms, as well as iOS and Android mobile.

macOS users can install it directly via Homebrew:

After installation, open Orca, connect your frequently used agents, and you're ready to go—the learning curve is very low.

04 Final Thoughts

Simply put, it helps developers solve the problem of managing multiple agents.

Previously, running multiple agents at once meant opening many terminal windows and switching back and forth—very troublesome.

Orca puts them all in one interface, and each agent has its own worktree, so they don't interfere with each other.

However, there are still some limitations: only CLI agents are supported, and GUI-based agents cannot be connected yet.

The mobile Companion feature is still being improved, but it's already sufficient for checking progress and remotely dispatching tasks.

Overall, the experience is quite good.

If you're interested, you can install it and give it a try.

图片

图片

图片

图片

图片

图片

图片

图片

图片

图片

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