We believe that everyone doing manual graphing work on a daily basis has been bothered by issues such as constantly having to adjust and pull the canvas, and images becoming outdated after code updates.
Using Mermaid to generate diagrams is convenient, but for more complex ones, the arrows may become disorganized, and manual adjustments are still necessary afterwards.
The open-source project Archify has recently topped GitHub Trending.
根据代码仓库的内容，生成一张高层运行时架构图如下： 1. 客户端（Client）：用户通过Web浏览器或移动应用程序访问系统。 2. 负载均衡器（Load Balancer）：将客户端请求分配到多个应用服务器，以确保系统的可用性和性能。 3. 应用服务器（Application Server）：运行业务逻辑代码，处理客户端请求，并与数据库交互。 4. 服务发现（Service Discovery）：管理应用服务器的实例，并提供服务注册和发现功能。 5. 数据库（Database）：存储系统的数据，包括用户信息、业务数据等。 6. 消息队列（Message Queue）：处理异步任务和消息通知，确保系统的可扩展性和可靠性。 7. 缓存层（Cache Layer）：缓存频繁访问的数据，减少数据库的负载，提高系统的性能。 8. API网关（API Gateway）：管理API的访问，提供安全、认证和限流功能。 运行时架构图： ``` +---------------+ | 客户端 | +---------------+
Then, an HTML-format architecture diagram that can search for nodes, track call paths, and switch themes is generated.
Moreover, it is model-agnostic, and can run regardless of whether you are using Claude Code, Codex, Cursor, or any other Agent.
As of press time, Archify has garnered numerous positive reviews with over 100,000 likes on X, and has accumulated 1.9k Forks and 31.4k Stars, propelling it to the top of GitHub's trending list.
Everyone can't help but exclaim:
Thanks to Archify, I've saved at least four or five follow-up meetings.
But surprisingly, the developer of this popular AI Agent "also no wind or rain or fog or sunshine" is not a graduate of a prestigious university.
He attended a junior college, and later was admitted to Chongqing University of Posts and Telecommunications' software engineering program as the top scorer in the specialist-to-bachelor's degree examination.
After Archify went viral, he joked somewhat wistfully on his X account:
I've worked as a security guard, studied car repairs, and even worked in a factory. I also managed to make it through the rigorous interviews at top companies like ByteDance's Feishu and Yuanfudao, only to be rejected during the background check because I had attended a junior college. The year I graduated, this happened to me repeatedly - I would have a great conversation during the interview, but as soon as I mentioned that I was a "junior college transfer student", I would never hear back from the HR department again.
He admitted that after one offer after another fell through, he had been in pain, questioned himself, felt lost, and even immediately went downstairs to run 5 kilometers.
But he never gave up.
Later, under the encouragement of Manus COO Cz Chen, he chose to use Archify to prove himself to everyone:
Humble beginnings are nothing to be ashamed of!
In one sentence: turn code into a system map.
Archify is an Agent Skill for AI intelligences such as Claude Code, Codex, Cursor, and OpenCode.
You can provide it with a system description, or let Archify directly analyze your code repository.
After understanding the code, Archify first generates a structured JSON IR, and then Archify completes schema, layout, and rendering checks, ultimately outputting a standalone HTML file.
The entire process is roughly as follows:
Code or text description → Archify understanding → JSON IR generation → rule validation → HTML/SVG rendering → interaction and export.
It currently can generate five types of technical diagrams: architecture diagrams, workflow diagrams, sequence diagrams, data flow diagrams, and lifecycle diagrams.
Moreover, the generated images are not just for viewing; you can also use them to search for nodes, view the upstream and downstream of a particular module, and track the path between two nodes. Additionally, you can switch between light and dark themes at will, export images in PNG, SVG, and WebM formats, and share them as cards.
If you are not satisfied with the results, you can also instruct Archify to make modifications at any time, for example:
Join Redis.
Move the authentication module to the left.
Highlight the fallback path after cache invalidation.
But the question arises, since tools like Mermaid and D2 for generating diagrams from text already exist, what makes Archify unique?
The answer is, it's not just about generating images from text, but rather embedding the drawing capability directly into the programming agent's workflow, allowing the agent to understand code, generate GIF-format charts, and continuously modify them.
By contrast, Mermaid and Graphify primarily focus on rendering existing text descriptions into images, and their output may not even be as polished as Archify's.
It's not just a prettier Mermaid
What truly sets Archify apart is that it embeds drawing tools into the workflow of its programming agents.
You don't need to learn the drawing grammar from scratch, just describe the technical intention.
While understanding code, organizing components, and generating structured data can be left to the AI Agent, Archify is responsible for verification and rendering.
This division of labor is also attempting to solve a problem with AI-generated architecture diagrams, namely
" I'm extremely happy to be a a of top of a only place of only top of minimal data of place of more common place of only place of a more common place of a only top of a only firsts of a more common place of a only top of a. I only place of a only place of a only place of a only place. " I only place of a. "I only place of a. " " Only place of only place of only. " " only place of only place of only. -," " only place of only place of only. " " only place of only place of only. " " only of top of a only place of only place of common. " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " only " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " "
Archify will check the data structure, node layout, labels, and wiring conflicts before delivery.
If verification fails, it will indicate specific problems and allow for corrective measures to be taken, rather than directly delivering a seemingly beautiful but actually flawed diagram.
Additionally, it can also compare two architecture snapshots, highlighting new, deleted, modified, moved, and rerouted parts in a Before/Delta/After manner.
This means that architecture diagrams are no longer just illustrations added after development is completed, but can also participate in code reviews and evolve with the project.
Of course, Archify is not flawless, it is not a real-time monitoring system and cannot automatically prove that the graph and the online running status are completely consistent.
The accuracy of the results still depends on the code information, user prompts, and the analytical capabilities of the Agent.
Some netizens who conducted tests also pointed out that Archify has issues such as the details display of generated nodes being too small, being somewhat inconvenient when using models without built-in visualization (such as deepseek-v4-flash), and consuming quota excessively.
In response, the developer "也无风雨也雾晴" stated that they are working hard to update the version and overcome these issues, and we can expect a wave of improvements.
So, Archify is suitable for someone like you:
Already utilizing AI programming, with the goal of quickly understanding unfamiliar repositories, sorting out call chains, or clearly explaining complex systems to the team.
But Archify's story is not just about how a tool can quickly go viral in four months.
Behind it lies a more specific issue:
When a degree can no longer be changed, how can a developer prove themselves?
The answer given by "also no wind or rain or fog or clear" is to make capabilities into works that anyone can open, use, and verify.
I'm a junior college student transferring to a four-year university, but I can do it!
The teacher, who goes by the pseudonym "也无风雨也雾晴", has not publicly disclosed his real name or his current company, but Archify's story has become increasingly clear and has won recognition from more and more people.
Now that the clouds have parted and the sky is clear, let's affectionately refer to him as "Teacher Sunny" from now on. (We've already been in touch with him, and interested friends can look forward to a follow-up in-depth interview).
Qingtian Teacher shared his story on social media, sincerely interacting with everyone and candidly sharing his journey from a specialist-to-bachelor front-end graduate, who faced numerous setbacks, to ultimately landing a job offer from a major company in Shanghai with his open-source projects and tools, with key points including:
Starting from rural areas, majoring in a specialized field and then pursuing a bachelor's degree, self-studying front-end development, repeatedly being held back by background checks, and open-sourcing to prove one's abilities.
After the 2018 college entrance examination, Teacher Qingtian went on to study computer application technology at a vocational school, a year during which he felt his condition was terrible.
Later, he was admitted to the software engineering program at Chongqing University of Posts and Telecommunications, ranking first in his cohort through the college-to-university upgrade pathway.
During his internship, Teacher Qing Tian was very happy, he felt:
As long as I keep working hard, the future will definitely get better and better.
But the autumn recruitment season poured a bucket of cold water on him:
During an internship interview at Westlake University, when the HR representative once again made an issue of his diploma from a junior college that he had used to pursue further studies, he finally couldn't help but explode in anger:
He finally passed.
But this was only an internship. Throughout Teacher Qingtian's job search, no company was ever willing to extend him an offer for a full-time position.
From last year up until this March, he had no choice but to constantly move from one internship to another, accumulating experience along the way:
Later, he finally met the first benefactor of his career - Cz Chen, who was then the Chief Operating Officer of Manus.
The diagram is sourced from the official Manus team directory △
She suggested that he could prove his abilities through his work or open-source projects, perhaps from other places.
This sentence not only confirmed the idea that Qingtian had been harboring in his heart all along:
Educational credentials are set in stone at this point—no matter how you explain it, what others see is still the 专升本 label.
It also became a foreshadowing thread in the later Archify story.
So he embarked on the path of open source, turning his work into a second resume along the way:
In the end, he relied on his own efforts and landed a job offer from a major manufacturer as he had hoped.
Later on, Archify became popular, and he was finally able to take this opportunity to share with everyone the discontent and confusion that had been deeply buried in his heart:
Open source didn't turn me from a junior college graduate into a 985 university graduate, and it can't guarantee that my education background won't hold me back again in the future. I still can't say clearly what those stars will ultimately bring me.
But in the end, he sighed:
At least for now, Archify is indeed being used by people, and the things I've created can bring value to others and help many people, which to me is the greatest significance of doing open source.
Beyond his academic credentials, he has already built for himself a new resume written through real work, user engagement, and community feedback.
His story offers hope to the many students who have found themselves shut out time and again for lack of academic qualifications.
And the story of Archify is still unfolding.
Reference links: [1]https://github.com/tt-a1i/archify [2]https://x.com/t20000622yy [3]https://trendshift.io/repositories/31352?utm_source=chatgpt.com
According to Manus AI's company profile on The Org, the company can be found at https://theorg.com/org/manus-ai. Additionally, a post on X by user t20000622yy mentions Manus AI in a status update, available at https://x.com/t20000622yy/status/2077425261856096720.