邀请你免费体验LobsterAI 有道龙虾
我的专属邀请码: cJdFVG
你也可以或者300积分。
立即下载: https://lobsterai.youdao.com/#/download-list?market=invitation
使用步骤:
1、点击上方链接下载 LobsterAI
2、安装打开应用,点击【登录】
3、登录页面输入邀请码 cJdFVG 即可获得 300 积分新人礼包
我已经在用了,真的很好用!
---
最近这段时间,Agent 可以说是 AI 圈最热的话题之一。
从代码开发到办公自动化,再到各种桌面 Agent,越来越多的人开始把 Agent 当成真正的工作伙伴,而不仅仅是一个 AI 对话工具。
这段时间,我也一直在用网易有道 LobsterAI 做一些工作和生活里的小项目。它是一个开源桌面 Agent,用户下载安装就可以直接使用,如果你本身就是开发者,也可以基于开源版本继续研究和扩展。
我比较喜欢它的一点,就是它更像是在围绕一个项目工作,而不是一次回答。整个项目都会保留在本地,想到新的需求可以继续在原来的基础上修改;做完之后,还能直接部署成一个可以访问的网站,生成分享链接,别人不用安装任何环境,打开浏览器就能直接体验。对于一些真正需要交付的项目来说,这种工作方式确实方便了不少。

最近,我就陆续用它完成了三个小项目:工作中会用到的安全事件管理系统、世界杯期间做的一个世界杯看板,还有孩子做的一个塔防小游戏。
1 安全事件管理系统
我是做网络安全运营的,平时经常要分析告警、处置安全事件、跟踪漏洞整改。以前这些数据大多记录在 Excel 表格里,当然也能用,但时间一长就会有几个问题:事件状态不够直观,处理进度不够清晰,统计起来也不够方便。尤其是给别人展示的时候,一张表格总归还是有点台账的感觉,不像一个真正的管理系统。
所以我一开始的目标就很明确:做一个安全事件管理系统。它不用像商业化产品那么复杂,但至少要比 Excel 更直观一些,可以把安全事件、严重等级、处理状态、负责人、事件详情这些信息集中管理起来,也能通过仪表盘看到待处理事件、高危事件等关键数据。
我把需求整理了一下,直接交给了LobsterAI。
设置目标:帮我开发一个专业的安全事件管理系统。
具体要求如下:
1.系统架构:必须搭建 Node 服务类站点,并结合数据库实现状态持久化,确保我添加、修改或删除的安全事件在页面刷新或软件重启后不会丢失。
2.核心功能(CRUD增删改查):
·【增】提供一个“新建事件"的表单或弹窗,字段包括:事件名称、严重程度(高、中、低)、发生时间、当前状态(待处理、处理中、已解决)、处理人、事件详情描述。
-【查】主界面为一个清晰的“安全事件列表看板”,能够直观看到所有事件的信息。·【改】在列表的每一行提供“修改“或“变更状态"的按钮,方便随时更新事件进展。·【删】提供“删除“或“归档“操作,可以移除误报或失效的事件。
3.界面设计:
-整体UI风格要偏向网络安全、科技感(可以采用深色调,如深海蓝或午夜深蓝)。
-页面顶部最好带一个简单的“实时数据仪表盘”,展示当前“待处理“和“高危“事件的数量统计。

这个是生成的本地源码,已经是一个小项目了。

创建分享,这次都是模拟数据我就用公开访问了,实际应用肯定还是得设置分享码的。

开发完成后,我直接点击了提交部署。
这也是 LobsterAI 最近新增支持的一项能力。像安全事件管理系统这种带有 Node 服务和数据持久化能力的项目,以前部署起来并不轻松。
最近不少 AI Coding 产品都开始提供 Sites 部署能力。Codex 的 Sites 目前主要以预览形式向 Enterprise 用户开放,而 LobsterAI 已经支持普通订阅用户使用,同时支持静态页面和 Node 服务类站点。
提交部署之后,它会自动完成网站部署,并生成一个可以访问的分享链接。我几乎不用再关心服务器配置、Nginx 或端口映射这些事情,部署完成后直接把链接发给同事体验就可以了。

看到部署状态完成即可。

第一版出来我觉得还是比较满意的,效果如下,可以直观展示安全事件和处理状态。

支持新建安全事件。

下面已经有刚才新建的挖矿木马事件了。

安全事件肯定是要支持批量导入功能的,我又加了新的需求。
必须支持批量导入功能,这样历史数据就不用一条条手工添加了。

可以导入多个安全事件了。

2 世界杯看板
7 月4号那天,我起得比较早,刚好看了半场阿根廷对佛得角的比赛。
世界杯总有这样的魅力,一场比赛,就能让你记住一个原本陌生的国家。
看完之后,我对这次世界杯的赛况产生了更浓的兴趣,干脆想做一个自己的世界杯看板,把积分榜、淘汰赛、射手榜这些信息都放在一个页面里。
这个需求本身不复杂,但有一个关键点:数据不能是模型随便生成的示例数据,而是要从网上抓取真实赛况。
我给LobsterAI的提示词里,明确要求它做一个 2026 世界杯数据展示网页,暗色足球主题,包含 12 个小组积分榜、淘汰赛对阵图、进球榜和助攻榜,数据从网络上抓取真实成绩。

调用web-search的skill抓取页面,直接操作自己本地浏览器,我看到数据基本源于下面的网站。

第一次效果已经基本成型。

可以对自己的网站进行实时修改,比如语言全部转换为中文,更新最新数据等。
赛况更新成7月6日的数据了,语言也更换成中文。

完成之后可以在LobsterAI浏览器打开,也可以在本地谷歌、Edge等浏览器打开。
点击右侧的分享按钮,可以生成链接,发送给其他人访问。

支持公开访问,或者设置分享码。值得注意的是,当你更新了需求,更新本地HTML文件后,如果你想同时更新在线版网页,需要点击一个更新文件。

这个小需求做完之后,我对它的第一印象还不错。它不是只帮我生成了一个静态网页,而是把找数据、做页面、更新内容、部署分享这一整条链路跑通了。
对于这种临时想到的小网页来说,最后能直接发一个链接给别人看,比单纯拿到一段代码更有意义。
3 双线塔防小游戏
这个小案例,是我孩子想出来的。
刚好他中考结束,这段时间比较喜欢塔防游戏,就自己设计了一套规则。塔防游戏大家应该都不陌生,简单来说,就是怪物沿着固定路线往基地走,玩家不能直接攻击怪物,只能在路线旁边建造防御塔,通过不同防御塔的搭配,尽量在怪物到达基地之前把它们消灭掉。
他这次提的需求还挺具体,不只是做一个塔防游戏这么简单,而是要有双线进攻、网格地图、不同价格和攻击方式的防御塔,还要有普通怪、高速怪、肉盾、飞行怪和后期 Boss。金币、血量、波次、塔的数量上限也都要有,整体难度还不能太低,要真的需要动点脑筋才能通关。
我让他把这些规则整理成提示词,直接交给了LobsterAI。

生成出来之后,我们先在本地玩了一会儿。整体完成度比我预想中高,怪物会沿着两条路线移动,防御塔可以自动索敌攻击,金币和基地血量也会实时变化。虽然肯定不能和成熟游戏比,但作为一个临时想出来的小项目,已经能玩起来了。

更有意思的是,游戏做好之后,孩子第一反应不是让我继续改,而是问:这个能发给同学玩吗?我点了一下分享按钮,生成了一个链接,他就直接发给同学了。对方不用安装任何环境,也不用下载代码,打开网页就能玩。
我觉得,这个案例最有意思的地方,不是游戏本身,而是它很快就完成了从自己玩到大家一起玩的过程。做一个自己的小游戏,再发给同学一起玩,这件事本身,是不是就挺酷的。
写在最后
Agent 发展到今天,大家关注的已经不只是模型能力。
真正值得关注的,是谁能把开发变成交付,把一个想法真正变成一个可以使用的项目。
网易 LobsterAI 给我的感觉,就是在努力做好这最后一公里。
I invite you to try LobsterAI for free. My exclusive invitation code: cJdFVG
You can also earn 300 points. Download now: https://lobsterai.youdao.com/#/download-list?market=invitation Steps: 1. Click the link above to download LobsterAI 2. Install and open the app, then click [Login] 3. On the login page, enter the invitation code cJdFVG to get the 300 points newbie gift pack I'm already using it and it's truly excellent!
---
Recently, agents have been one of the most popular topics in the AI community.
From code development to office automation, and various desktop agents, more and more people are starting to use agents as true work partners, not just AI chat tools.
During this time, I've been using NetEase's LobsterAI to handle some work and life projects. It is an open-source desktop agent that users can download and install for immediate use. If you are a developer, you can also further research and extend from the open-source version.
One thing I like about it is that it operates more like a project than a one-time response. The entire project is retained locally; you can continue modifying it based on new requirements, and once finished, you can directly deploy it as a website. You can generate a share link, allowing others to experience it in their browser without needing to install any environment. For projects that require actual delivery, this workflow is significantly more convenient.
Recently, I've completed three small projects: a security event management system for work, a World Cup dashboard during the tournament, and a tower defense game for my child.
1 Security Event Management System
I work in network security operations and frequently need to analyze alerts, handle security events, and track vulnerability remediations. In the past, this data was mostly recorded in Excel sheets. While functional, it has several drawbacks over time: the event status is not intuitive, progress is unclear, and statistics are cumbersome. When presenting to others, a spreadsheet feels like a ledger rather than a proper management system.
So, my initial goal was clear: to create a security event management system. It wouldn't need to be as complex as a commercial product, but it should be more intuitive than Excel. It should centralize information about security events, severity levels, handling statuses, and responsible personnel, and provide a dashboard for key data like pending events and high-risk occurrences.
I summarized my requirements and fed them into LobsterAI.
Set the goal: Develop a professional security event management system.
Specific requirements:
1. System architecture: Must build a Node service site combined with a database for state persistence, ensuring that added, modified, or deleted events are not lost after a page refresh or software restart.
2. Core features (CRUD - Create, Read, Update, Delete):
- [Create] Provide a form or pop-up for "new event" with fields: event name, severity (high, medium, low), occurrence time, current status (pending, in progress, resolved), handler, and event description.
- [Read] The main interface should display a clear "security event list dashboard" showing all event information.
- [Update] Each row should have a button for "modify" or "change status" to easily update progress.
- [Delete] Provide "delete" or "archive" options to remove false positives or expired events.
3. User interface:
- A network security, tech-friendly style (dark tones like deep blue or midnight blue).
- A simple "real-time data dashboard" at the top to show counts of pending and high-risk events.
This is the generated local source code, already set up as a small project.
I created a share link, using public access for these simulation data. In a real application, I'd set a share code.
After development, I clicked submit for deployment.
This is a recent feature added to LobsterAI. Projects like security event management systems, which involve Node services and data persistence, were previously difficult to deploy.
Recently, many AI coding tools have started offering Sites deployment. Codex's Sites are currently available to Enterprise users as a preview, while LobsterAI supports standard subscription users and handles both static pages and Node service sites.
This is the generated local source code, and it is already a small project.
Creating a share, since these are all simulated data, I will use public access; in actual applications, a share code will certainly need to be set.
After development, I directly clicked on submit and deploy.
This is also a feature that LobsterAI has recently added support for. Previously, deploying a project like a security event management system, which includes a Node service and data persistence capabilities, was not easy.
Recently, many AI coding products have begun to provide Sites deployment capabilities. Codex's Sites are currently open to Enterprise users primarily in preview form, while LobsterAI already supports ordinary subscribers to use it, and supports both static pages and Node service sites.
Once deployed, it automatically completes the website setup and generates a shareable link. I no longer need to worry about server configurations, Nginx, or port mappings—just share the link with colleagues to get started.
The deployment status shows as “Completed”, indicating the process is successful.
I’m quite satisfied with the first result—the dashboard clearly visualizes security events and their handling status.
New security events can be easily created from the interface.
The system already shows the newly added mining malware event.
Since I plan to import many historical events, I requested a feature for bulk importation.
Bulk import is essential to avoid manually adding every single event.
Now, multiple security events can be imported at once.
2. World Cup Dashboard
On July 4th, I woke up early and watched the first half of Argentina vs. Cape Verde.
World cups always reveal the beauty of lesser-known countries through soccer. Inspired by this, I decided to build my own dashboard to track scores, knockouts, and top players.
The requirement was simple but critical: data must not be random placeholders—it should be fetched from real-world sources.
I prompted LobsterAI to create a 2026 World Cup data dashboard with a dark soccer theme, including 12 group tables, knockout brackets, goals, and assists lists, all populated by scraping official sites.
The final product can be viewed in LobsterAI's built-in browser or exported to Google Chrome, Edge, etc.
After completion, the file can be opened in the LobsterAI browser or any local browser like Google Chrome or Edge.
You can click the share button on the right to generate a link and send it to others for access.
You can choose to make it publicly accessible or set a share code. Importantly, if you update your requirements and the local HTML file, you need to click "Update File" to synchronize the online page.
After completing this small project, my first impression was quite positive. It didn't just generate a static webpage; it completed the entire pipeline from data fetching to page creation, content updating, and deployment for sharing.
For temporary web projects like this, being able to share a direct link immediately is much more valuable than just receiving a code snippet.
3. Two-Lane Tower Defense Game
This specific project idea came from my child.
He recently finished his middle school exams and has been enjoying tower defense games. He decided to design his own set of rules. The standard tower defense premise—monsters move toward the base, and players build towers to stop them—is well-known, but he added specific constraints: a grid map, distinct tower types with varying costs and attacks, and a diverse enemy lineup including fast, tanky, flying, and boss-level monsters. He emphasized that the difficulty should be high enough to require strategic thinking.
I had him write down these requirements in a prompt and fed it to LobsterAI.
The most interesting aspect of this case isn't necessarily the game itself, but how quickly it transitioned from a personal project to a shared experience. The ability to instantly publish and share a custom game is what really stands out as "cool."
Final Thoughts
As Agent technology evolves, the focus is shifting from raw model capacity to the practical application of those models.
What really matters is who can bridge the gap between development and delivery, turning a simple idea into a usable, shareable product.
NetEase LobsterAI appears to be making significant strides in covering this "last mile," effectively turning prototypes into playable experiences.
To me, NetEase LobsterAI feels like it's working hard to bridge the last mile.