Source: https://www.youtube.com/watch?v=daXOXSSyudM
最近几天很多粉丝让我讲讲最近爆火的CloudBot。这款开源的个人AI助手项目。之所以这个项目能火是因为它是第一个真正实用的自拖管AI员工。本质上就是一个本地运行的AI智能体框架。经过这几天的测试发现CloudBot的功能非常强大。而且应用场景非常广泛。甚至感觉CloudBot越用越聪明。因为它能通过长期的聊天实现自我进化。能够记住用户的偏好还能主动提供建议。而且CloudBot的能力不是固定的。我们可以在CloudBot中通过安装不同的skills。让CloudBot具备不同的能力。所以如果将CloudBot运用好。它能大幅度提升我们的工作和学习效率。而且它的部署非常简单。只需要一条面链五到十分钟就可以完成安装。而且我们可以将这个项目部署到多种平台。比如说micOS系统 原因服务器。甚至是树莓派。而且不一定非要买Mac系统的电脑。我们甚至可以将旧电脑或者旧笔记本上。装上Linux系统来部署这个项目。而且它还能主动执行任务。具备无限的记忆远超普通的聊天AI。最关键的是我们可以通过WaltzApp。Basecard等聊天App进行交互。我们可以通过实习的聊天方式。让AI帮我们做任何事情。无论是控制电脑 实现自动化工作流。还是开发应用。它把AI聊天工具变成了真正的执行者。本期视频将先为大家演示CloudBot的布什方式。然后我们结合几个比较实用的案例。来测试CloudBot它的综合能力到底怎么样。在演示之前。我们可以先看一下CloudBot它的系统架构。首先是用户层。用户可以通过WaltzApp等及时通讯工具。实现于CloudBot进行交互。然后第二层就是区道层。在区道层能够实现协议适配。消息解析格式转换。媒体处理分块传输。然后就到了网关核心。在网关这里可以实现汇化管理。消息路由工具调用自动化。它具备的核心功能。包括多区道统一受电箱。还有浏览器自动化。还有系统级的完整防卫。而且可以实现与音换型与对话。还能实现可实化工作区。还具备定时任务自动化功能。而且它具备成熟的skills生态。并且支持多只能体协作。下面为大家想显示。我们如何在本地部署这个项目。首先我们直接复制官方文档中。给出的这条命令。然后打开中端命令行。我们直接将命令粘贴到中端命令行中。直接运行就可以。到这一步的时候。要选第一个选项。要同意这个协议。当同意协议之后。这里需要我们设置一下模型的提供商。在模型提供商这里。大家如果有OpenAI的订阅。那么可以直接使用OpenAI的。Codex订阅进行登录。如果想使用API Key的话。也可以直接选择API Key。第二项是安索奥匹克。大家如果订阅了Cloud。也可以直接通过Cloud订阅进行登录。然后到这一步的时候。我们就可以选择使用哪一种聊天工具。与CloudBot进行交互。它支持多种聊天工具。这里我就选择WhatsApp。然后我们直接选中。这里会在中端命令行出现一个二维码。然后我们需要在手机上打开WhatsApp。扫描这个二维码实现设备的连接。当这一些完成之后。下面我们需要设置一下。它调用哪些技能。在这里需要我们选择技能的安装方式。我这里就选择第一种。然后这里就会出现多种skills。让我们去安装。大家可以根据自己的需要去安装。也可以直接选择跳过当前的这些安装。然后这里其实需要设置这些API Key。然后这里我们可以先略过。到这一步就是取用Hooks。我这里可以选择第四项。让它实现这个Session的记忆。当这些设置好处。我们直接重新运行就可以。然后到这一步。我就默认选择第一项。这样的话。它就自动在我浏览器中。打开了CloudBot的这个后台管理的页面。这里有一个类似于拆的GPG的对话框。在输入框中。我们就可以输入一个内容测试一下。在这里我让它讲个故事。然后这里它就输出了。它讲了一个故事。它不仅支持直接在网页后台进行对话。还支持刚才我们连接好的WhatsApp进行对话。比如说我们在WhatsApp中让它讲个故事。然后直接发送。这里它很快。会有我们讲的一个故事。在这里我们还可以点击阅读更多。这样的话我们就可以完整的来查看这个故事。在WhatsApp中。我们还可以查看我之前使用CloudBot。来执行的这个任务。比如说这里自动让它抓取某些网站。或者博客上的内容。当抓取完成之后。它就会自动推送到WhatsApp。下面我们先在WhatsApp中进行交互。来测试一下CloudBot。它的浏览器自动化能力。我这里属于的提示词时。让它调用浏览器来打开CloudBot的官方仓库。并给出这个项目的安装命令。然后我们直接发送。可以看到这里它自动打开了浏览器。并且在浏览器中打开了CloudBot官方仓库。然后我们看一下。它能否输出这个项目的安装方式。它很快输出了这个项目的安装方式。包括推荐的安装方式是用APM全局安装。这里就给出了具体的安装命令。还有启动命令。还有快速测试。像这样的话。我们就实现了在WallTap中直接通过聊天的形式。让CloudBot为我们执行了浏览器自动化任务。像这样的话。哪怕我们不在电脑旁边。也可以通过手机上安装的WallTap。来操控电脑上的CloudBot为我们执行各种复杂的任务。在这个网页版的管理后台。大家就可以根据自己的需求来安装对应的skills。也就是我们需要CloudBot为我们完成哪些任务。我们就可以安装对应的skills。像安装这些skills非常简单。我们只需要在右侧点击对应的安装就可以。比如说我这里安装了BlogWater。它可以监控各种博客是否发布了最新的更新。安装好之后CloudBot就可以来调用我们安装的这些skills。下面我们就可以通过CloudBot它的定时任务。来调用刚才我们安装的BlogWater。每天定时为我们执行抓取相关的技术文章。这样的话我们每天就可以准时在WallTap上查看。指定网站或者博客的文章更新。我们就可以在CloudBot后台来设置定时任务。而且还可以通过命令的方式来设置定时任务。为了快损言识。我们可以直接通过命令的方式来实现创建定时任务。因为刚才我们添加了BlogWater这个skills。所以我们就可以通过BlogWater的命令。将我们需要查看的一些博客添加到BlogWater中。添加的时候非常简单。我们直接打开中断命令。然后直接执行刚才我们查看的命令。下面我们就可以来执行这一条命令。这条命令的功能就是每天9点。来检查这些博客是否有更新。如果有更新就推送到WallTap。在名称这里就是这个任务的名称。然后这里就是设置的每天9点。这里是时区。然后大家可以根据自己的所在地来修改时区。然后这里就是给他设置的任务的提示词。要求大家使用BlogWater这个工具。来扫描这些订阅。并且列出这些最新的文章。确保只抓取于AI大模型或者Agent。或者编程工具相关的这些内容。然后在这个参数这里。就让他退送到我的WallTap。下面这个参数就是我的WallTap的这个电话号码。然后我们就可以完整的复制这条命令。在中断命令行中。我们直接粘贴直接运行。这条命令就可以。好这个命令执行成功。然后我们回到CloudBot后台的定时任务这里。我们直接刷新一下。这里就看到了我刚才用命令创建的定时任务。然后在右侧我们就可以点击运行。让他立即执行这个任务。然后我们好看一下效果。我们直接点击执行。点击执行之后。在下面这里就会显示这个运行的历史。这里就释出了这个今日AI简报。这里提示当前订阅员。仅补货10篇一都文章。全部来自我的博客。好下面我们就可以回到WallTap中。来查看一下。他是否将刚才的这个消息推送到了WallTap。在WallTap中。我们可以看到这里他已经将刚才的消息。自动推送到了我们的WallTap。因为这里我博客上的10篇文章。我在之前已经阅读完毕。所以这里他就提示今天没有可更新的。这些文章。这是我们测试的在CloudBot中。通过创建定时任务。来实现将定时任务执行的结果。推送到WallTap上的工作流。下面我们还可以继续测试。我们还是先点击Skills。然后我们找到CodingAgent和JigSkill。JigSkill的功能就是。它的运行CodeXCLI CloudCode OpenCode。通过后台进程的方式来实现程序控制。下面我们就可以在WallTap中。通过JigSkill为我们编写代码。下面我们就可以在WallTap中试试试试试。让它使用CodingAgent这个Skill。调用CloudCode开发一个后台登录页。并调用浏览器查看效果。我们直接运行查看一下效果。再等待了两分钟左右。这里他自动在浏览器中打开了。为我们开发的这个后台登录页。可以看到它开发的这个后台登录页。效果还是非常不错的。这是我们测试的让它为我们进行。编程开发的行为。它能自动调用浏览器。来打开为我们开发好的这个登录页。像这样的话。当我们不在电脑前的时候。就可以通过手机WallTap。来操控CloudBot为我们进行编程开发。而且CloudBot它支持的JigSkill非常非常多。由于时间有限。本期视频只为大家演示了。基础的使用方式。后续视频我还会为大家演示。CloudBot的高级用法。好本期视频。所用到的代码后指令。我都会放在视频下方的描述栏。或者评论区。如果你在视频下方无法找到的话。也可以通过我的博客。去查找本期视频所对应的笔记。好本期视频就做到这里。欢迎大家点赞关注和转发。谢谢大家观看。
In recent days, many fans have asked me to talk about the recently popular CloudBot, an open-source personal AI assistant project. The reason this project has become so popular is that it's the first truly practical self-hosted AI employee. Essentially, it's a locally running AI agent framework. After testing it for the past few days, I've found CloudBot to be incredibly powerful with a wide range of applications. I even feel that CloudBot gets smarter the more I use it, as it can self-evolve through long-term conversations. It can remember user preferences and proactively offer suggestions. Furthermore, CloudBot's capabilities aren't fixed; we can equip it with different abilities by installing various skills within CloudBot. Therefore, if utilized effectively, it can significantly boost our work and study efficiency. Its deployment is also very simple, requiring just one command and five to ten minutes to complete the installation. Moreover, we can deploy this project on various platforms, such as macOS, cloud servers, or even a Raspberry Pi. You don't even need to buy a Mac; you can deploy this project on an old computer or laptop by installing a Linux system. It can also proactively execute tasks and possesses unlimited memory, far surpassing ordinary conversational AIs. Most importantly, we can interact with it through chat apps like WaltzApp and Basecard. We can use a conversational approach to have the AI help us with anything, whether it's controlling a computer, implementing automated workflows, or developing applications. It transforms AI chat tools into true executors. This video will first demonstrate CloudBot's deployment method, and then we'll test its comprehensive capabilities with a few practical examples. Before the demonstration, let's take a look at CloudBot's system architecture. First, there's the user layer, where users can interact with CloudBot via instant messaging tools like WaltzApp. The second layer is the channel layer, which handles protocol adaptation, message parsing, format conversion, media processing, and chunked transfer. Next is the gateway core, where session management, message routing, and automated tool invocation are handled. Its core functionalities include a multi-channel unified inbox, browser automation, and system-level comprehensive defense. It also supports voice interaction and conversation, provides a visual workspace, and features automated scheduled tasks. Furthermore, it boasts a mature skills ecosystem and supports multi-agent collaboration. Next, I'll show you how to deploy this project locally. First, we'll copy the command provided in the official documentation. Then, open the terminal command line, paste the command directly into it, and run it. At this step, select the first option to agree to the protocol. After agreeing to the protocol, we need to set up the model provider. For the model provider, if you have an OpenAI subscription, you can log in directly using your OpenAI Codex subscription. If you prefer to use an API Key, you can select that option directly. The second option is Anthropic; if you have a Cloud subscription, you can also log in directly using your Cloud subscription. At this point, we can choose which chat tool to use for interacting with CloudBot. It supports various chat tools; here, I'll choose WhatsApp. After selecting it, a QR code will appear in the terminal command line. Then, we need to open WhatsApp on our phone and scan this QR code to connect the device. Once these steps are complete, we need to set which skills it will invoke. Here, we need to choose the skill installation method; I'll select the first option. Then, various skills will appear for us to install. You can install them according to your needs, or you can choose to skip the current installations. Actually, API Keys need to be set here, but we can skip this for now. This step involves using Hooks. I'll select the fourth option to enable session memory. Once these settings are complete, we can simply restart it. Then, at this step, I'll select the first option by default. This will automatically open CloudBot's backend management page in my browser. There's a ChatGPT-like dialog box here. In the input field, we can enter some content to test it. Here, I asked it to tell a story, and it outputted one. It not only supports direct conversation in the web backend but also through the WhatsApp connection we just set up. For example, if we ask it to tell a story in WhatsApp and send it, it quickly provides a story. Here, we can also click "Read More" to view the complete story. In WhatsApp, we can also view tasks I previously executed using CloudBot. For instance, automatically fetching content from certain websites or blogs. Once fetched, it automatically pushes the content to WhatsApp. Next, let's interact with CloudBot in WhatsApp to test its browser automation capabilities. My prompt here is to have it use the browser to open CloudBot's official repository and provide the project's installation command. After sending it, we can see that it automatically opened the browser and navigated to CloudBot's official repository. Now let's see if it can output the project's installation method. It quickly outputted the project's installation method, including the recommended APM global installation. It provided the specific installation command, startup command, and quick test instructions. This way, we successfully had CloudBot perform a browser automation task for us directly through chat in WhatsApp. This means that even if we're not at our computer, we can control CloudBot on the PC via WhatsApp installed on our phone to perform various complex tasks. In this web-based management backend, you can install corresponding skills according to your needs. In other words, for any task we want CloudBot to complete, we can install the relevant skills. Installing these skills is very simple; we just need to click "Install" next to the desired skill on the right. For example, I've installed BlogWater, which can monitor various blogs for the latest updates. Once installed, CloudBot can invoke the skills we've set up. Next, we can use CloudBot's scheduled tasks to invoke the BlogWater skill we just installed, having it fetch relevant technical articles for us daily at a set time. This way, we can check for article updates from specified websites or blogs on WhatsApp punctually every day. We can set up scheduled tasks in the CloudBot backend, and we can also do so via commands. For a quick demonstration, we can create a scheduled task directly using a command. Since we just added the BlogWater skill, we can use BlogWater's command to add the blogs we want to monitor to BlogWater. Adding them is very simple: just open the terminal and execute the command we just reviewed. Now we can execute this command. Its function is to check these blogs for updates every day at 9 AM and push any updates to WhatsApp. "Name" here refers to the task's name, and "9 AM daily" is the set time. This is the time zone, which you can modify according to your location. And here is the prompt for the task, instructing it to use the BlogWater tool to scan these subscriptions and list the latest articles. It ensures that only content related to AI large models, agents, or programming tools is fetched. Then, in this parameter, it's set to push to my WhatsApp. The next parameter is my WhatsApp phone number. We can then copy this entire command, paste it directly into the terminal command line, and run it. Okay, the command executed successfully. Then, we'll go back to the scheduled tasks section in the CloudBot backend and refresh the page. Here, we can see the scheduled task I just created using the command. Then, on the right, we can click "Run" to have it execute this task immediately. Let's check the result. We'll click "Execute." After clicking "Execute," the execution history will be displayed below. Here, it released today's AI brief, stating that for the current subscriber, only 10 articles were retrieved, all from my blog. Okay, now we can go back to WhatsApp to check if it pushed the message there. In WhatsApp, we can see that it automatically pushed the message there. Since I had already read the 10 articles from my blog, it indicated that there were no new updates for these articles today. This is the workflow we tested in CloudBot: creating a scheduled task to push its execution results to WhatsApp. Next, we can continue testing. We'll first click on "Skills," then find "CodingAgent" and "JigSkill." JigSkill's function is to run CodeXCLI, CloudCode, and OpenCode, enabling program control through background processes. Now, we can use JigSkill in WhatsApp to write code for us. Next, in WhatsApp, we'll try using the CodingAgent skill to invoke CloudCode to develop a backend login page and then use the browser to view the result. We'll run it directly to see the effect. After waiting for about two minutes, it automatically opened the backend login page it developed for us in the browser. As you can see, the backend login page it developed looks quite good. This is our test of its programming development capabilities. It can automatically invoke the browser to open the login page it developed for us. This means that when we're not in front of our computer, we can control CloudBot on the PC via WhatsApp on our phone to perform programming development. CloudBot supports a vast number of JigSkills. Due to time constraints, this video only demonstrated the basic usage. In future videos, I will demonstrate CloudBot's advanced features. All the code and commands used in this video will be placed in the description box below the video or in the comment section. If you can't find them below the video, you can also check my blog for the notes corresponding to this video. That concludes this video. Please like, subscribe, and share. Thank you for watching.