Marvis 上手一周,我发现它跟 WorkBuddy、QClaw 完全不是一类东西After using Marvis for a week, I realized it’s fundamentally different from WorkBuddy and QClaw.


AI free token/AI免费积分

腾讯在五月底连甩两张牌,一张是 QClaw,另一张叫Marvis。

很多人第一反应是,又一个 AI 助手。但用了一周之后,我的判断变了。这三个东西虽然都叫"AI 助手",但它们解决的是完全不同的问题。

先说 Marvis 到底能干什么

Marvis 的定位很清楚,操作系统级的个人 AI 管家。

不是聊天框,不是插件,是直接住进系统底层的管家。

它能干什么

帮你清理 C 盘垃圾,关掉 Windows 广告弹窗,优化开机启动项。一句话的事,它直接调系统 API 去执行,不是告诉你"你可以去设置里怎么怎么点",而是它自己就去点了。

整理桌面文件,按人脸给照片分类,搜遍全盘 PDF 提取关键信息。本地文件零上传,全部在端侧完成。

手机远程操控电脑,锁屏状态下也能用。你在外面掏出手机就能看到电脑屏幕上正在执行什么任务,不满意随时接管。

还有个很有意思的设计,多 Agent 协同架构。一个主 Agent 当项目经理拆解任务,下面五个副 Agent 各管一摊。界面是个虚拟办公室,你能看到每个 Agent 是在搬砖还是在摸鱼。

两种运行模式随时切换。效率模式走云端,速度快精度高;隐私模式纯本地跑,断网也能用,文件不上传。

它跟 WorkBuddy 的区别

我用 WorkBuddy 很久了,写文章、跑脚本、管理公众号流水线,它是我每天离不开的生产力工具。

但 WorkBuddy 和 Marvis 走的是两条路。

WorkBuddy 本质上是开发者的 AI 工作空间。它的强项在于代码执行、文件操作、自动化流程编排,以及那个丰富的 Skill 生态。你在里面写代码、调 API、跑自动化任务,效率极高。

但它不碰你的操作系统。WorkBuddy 不知道你的电脑开了多少个高耗电进程,不知道你的 C 盘快满了,也不会帮你把桌面上的截图按项目归档。

Marvis 恰好补上了这块短板。

它是从系统层面切入的,能读硬件信息,能改系统设置,能管进程和启动项。这些事 WorkBuddy 干不了,也不该干,因为这不是一个开发者工具的职责范围。

反过来,如果你要跑一段 Python 脚本处理数据,或者用 Skill 生态完成一套复杂的自动化工作流,Marvis 目前也干不了这么精细。它的能力边界在系统操作和文件整理,不在代码开发和复杂逻辑编排。

所以我的实际用法是两个一起用。日常开发和内容生产走 WorkBuddy,电脑维护和文件管理交给 Marvis。各管各的赛道,互不冲突。

它跟 QClaw 的区别

这个对比更有意思,因为 QClaw 和 Marvis 都是腾讯出的,而且发布时间只隔了一个月左右。

QClaw 的核心卖点是微信直联。你发一条微信消息,它就在你电脑上干活,结果回传到微信里。不用装额外 App,不用开网页,手机就能远程指挥电脑。对非技术用户极其友好,"养龙虾"的概念也很讨巧。

但 QClaw 和 Marvis 的底层思路不一样。

QClaw 基于 OpenClaw 开源框架构建,本质是一个通过微信接入的本地 AI Agent。它的交互入口是微信,执行端是你电脑上的客户端。适合的场景是远程办公,人在外面需要电脑帮忙干活的时候。

Marvis 不依赖微信,它直接嵌入操作系统。这意味着它不需要额外的通信层,响应更快,而且能做到一些 QClaw 做不到的事情。比如锁屏状态下的远程操控,比如直接读取硬件信息和系统配置,比如真正意义上的系统级 API 调用。

隐私方面两者都能本地运行,但 Marvis 的本地模式做得更彻底,端侧大模型原生支持,不是简单的"不传云端"而是整个推理链路都在设备上完成。

还有一个差异点是场景覆盖面。QClaw 偏向办公场景,邮件、文档、代码、学术研究。Marvis 除了办公,还覆盖了追星、游戏、娱乐这些泛生活场景。它能帮粉丝做明星资讯监控和物料整理,能帮玩家做游戏每日任务的智能规划。这些东西 QClaw 没有,因为它的定位就是生产力工具。

简单打个比方。

QClaw 像一个靠谱的远程助理,你微信喊一声他就干活。Marvis 像一个住在你电脑里的全能管家,不光能干活,还懂你的电脑比你自己还清楚。WorkBuddy 则像一间装备齐全的工作室,工具齐全任你调用。

什么时候该选 Marvis

用了这段时间,我总结了几种特别适合用 Marvis 的场景。

电脑维护不想折腾的人。 清理垃圾、优化启动项、检查网络状态、关闭系统广告,这些事以前要么自己一个个去设置里翻,要么装各种清理软件担心有捆绑。现在一句话搞定,而且没有第三方软件的安全顾虑。

文件多到失控的人。 桌面堆满截图和下载文件,照片散落在各个文件夹找不到。Marvis 能按人像、内容主题、时间地点自动分类,还能建 AI 图库和文档库当个人知识库用。

经常出门但需要用到电脑的人。 手机远程控制电脑不是新概念,但锁屏状态下还能操作、实时看到执行画面、随时接管,这个体验确实比传统远控软件顺滑很多。

在意数据隐私的人。 本地模式全程断网可用,敏感文件处理不用担心泄露。对金融、法律、医疗这些行业的人来说,这个能力可能是决定性的。

一些真实的体感

说几个使用中的真实感受。

好的方面。系统级操作确实省心,尤其是文件整理功能,丢给它一堆乱七八糟的文件,它自动按类别归档,准确率比我预想的高。手机远程控制的延迟可以接受,日常查看进度和做简单操作没问题。虚拟办公室的多 Agent 可视化做得挺有意思,看着它们各自忙碌有一种奇怪的满足感。

需要改进的地方。Token 消耗确实比较大,复杂任务跑起来比较吃资源。执行速度有时候不够快,尤其是涉及多步骤的任务链。偶尔会偏离预期,目前还不能完全放手让它独立处理高复杂度的操作。

不过考虑到这是刚发布不到两周的产品,这些问题后续迭代应该会逐步改善。核心能力和产品定位已经立住了,剩下的就是打磨细节。

最后说句实在话

AI 助手这个赛道越来越卷了,但这其实是好事。WorkBuddy 管开发与自动化,QClaw 管微信远程办公,Marvis 管系统级智能管家。三者各有侧重,与其说是竞争关系,不如说是在把"AI 真正帮你干活"这件事往更深的方向推。

如果你还没试过 Marvis,值得花半小时体验一下。毕竟免费每天一千万 Token,试错成本几乎为零。

 

Tencent dropped two bombs at the end of May: QClaw and Marvis.

Most people's first reaction was, "Oh, just another AI assistant." But after a week of use, my opinion changed. Although all three are called "AI assistants," they solve completely different problems.

So what exactly can Marvis do?

Marvis is positioned as an operating system-level personal AI butler.

Not a chat window, not a plugin — it's a butler that lives directly in the underlying system.

Here's what it does:

Clean up C-drive junk, close Windows ad pop-ups, optimize startup items. Just one sentence, and it directly calls the system API to execute — it doesn't tell you "go to settings and click here," it clicks for you.

Organize desktop files, classify photos by face, search the entire drive for PDFs and extract key information. Local file processing with zero upload — everything happens on the device.

Remote control from phone to PC, even in lock-screen mode. You can see what your PC is doing from outside, and take control if you don't like the result.

It has a cool Multi-Agent architecture: a main agent coordinates the project, and five sub-agents handle specific tasks. You can see them working or slacking in a virtual office view.

Two running modes: Efficiency Mode (cloud-based, fast and accurate) and Privacy Mode (pure local, offline, no upload).

How is it different from WorkBuddy?

I've been using WorkBuddy for a long time — writing, running scripts, managing pipelines — it's my daily productivity tool.

But WorkBuddy and Marvis are on two different tracks.

WorkBuddy is essentially a developer's AI workspace. It shines in code execution, file operations, automation, and its rich Skill ecosystem. You write code, call APIs, run automated tasks — very efficient.

But it doesn't touch your operating system. WorkBuddy doesn't know if your C-drive is full, or which processes are running hot, or if you have unorganized screenshots.

Marvis fills that gap.

It starts from the system level — reads hardware info, changes settings, manages processes. These are things WorkBuddy doesn't do because it's not its job to be a system tuner.

Conversely, if you need to run a complex Python script or build a multi-step automation workflow, Marvis isn't there yet. Its focus is system operations and file management, not deep algorithmic development.

So my actual usage: WorkBuddy for development and content production; Marvis for PC maintenance and file organization. They don't conflict — they complement each other.

How is it different from QClaw?

This is interesting because both are from Tencent, and released almost simultaneously.

QClaw's key feature is Wechat integration: you text your PC, it does the work, and replies back. No extra app, no web login — remote control via phone using Wechat. Very user-friendly, especially for non-techies.

But QClaw and Marvis have different philosophies.

QClaw is based on OpenClaw, a local agent accessed via Wechat. It's a remote execution service tied to a messaging app. Good for office work when you're away.

Marvis doesn't rely on Wechat — it's built into the OS. This means faster response, deeper system access, and features like remote control while locked. It can truly be a "system-level" butler.

Privacy-wise, both offer local modes, but Marvis goes further with native client-side models, ensuring no data ever leaves the machine.

And Marvis covers more life scenarios — like fandom, gaming, entertainment — which QClaw doesn't target. QClaw is purely productivity; Marvis is more general-purpose.

To summarize:

QClaw is a reliable remote assistant you call via Wechat. Marvis is a live-in butler who knows your PC better than you. WorkBuddy is your fully equipped workshop for coding and automation.

When should I use Marvis?

After testing, I found Marvis particularly useful in these scenarios:

You hate fiddling with computer settings. Disk cleanup, startup optimization, network checks — Marvis handles it all.

You have too many files and no organization. Marvis sorts photos by face, groups docs by topic, creates libraries — all automatically.

You need remote access without hassle. Control your PC from your phone, even when the screen is off.

You care about data privacy. Local mode ensures no sensitive files go online.

Real-world impressions:

The good: System-level operations are a time saver; file organization is surprisingly accurate; the virtual office view is fun; phone control is responsive enough for most tasks.

The need for improvement: Token consumption is high for complex tasks; execution speed sometimes lags; occasional misinterpretation of instructions — but given it's only two weeks old, these are expected growing pains.

Final thoughts:

The AI assistant field is getting crowded, but that's good for us. WorkBuddy, QClaw, Marvis — each excels in a niche. Instead of competing, they're pushing the envelope on what "AI that actually helps" means.

If you haven't tried Marvis yet, give it a shot — 10 million free tokens per day make it risk-free to explore. And who wouldn't want a butler that cleans your PC while you make coffee?

The AI assistant race is getting hotter, but that's actually a good thing. WorkBuddy handles development and automation, QClaw manages remote office via WeChat, and Marvis is the system-level smart butler. Each has its own focus, and rather than competing, they're driving "AI actually helping you work" toward deeper integration.

If you haven't tried Marvis yet, it's worth half an hour of your time. After all, with free access to 10 million tokens per day, the cost of experimentation is effectively zero.

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