Skip to content
Before 2023, I thought API calls were exclusive to developers.
Until one day, I discovered a somewhat magical fact:
With the development of AI large models to this stage, I can also call APIs.
I can’t understand the documentation? Let AI handle it; I can’t write code? Let AI do the coding.
Recently, everyone has been discussing DeepSeek, and many people must have registered on DeepSeek’s open platform. After registration, DeepSeek’s open platform offers a $10 trial credit. It should not display the token count now, but before February 8, this $10 credit was equivalent to about 5 million tokens. Some may think that such a large number of tokens is excessive, but once you start using AI, you’ll find it barely suffices.
From the end of December last year until now, in less than a month, I’ve exhausted my trial credits, using about 9 million tokens in total for input and output. Let me share where I’ve used them:
The easiest application scenario is the browser extension, which is the immersive translation plugin I mentioned earlier: Too many English comments on Xiaohongshu that I can’t understand? I recommend trying the immersive translation.
Just fill in the API Key and endpoint address to use it.
Besides immersive translation, it can also organize and summarize page content.
For example, it helps you find your desired gender among a mixed list on a dating website:
Or, when you can’t wait to participate in such a dating event, you can quickly extract the headers from others’ profiles with a click and fill them in directly to sign up:
This way of using the browser extension is not difficult; anyone who can fill in forms can use it. However, it can easily be replaced by similar products from big companies, like Doubao.
The Doubao plugin is completely free, and after a few iterations, it can achieve the same functionality, while I’m here consuming API credits, which feels a bit unfair.
In fact, from the few functions of the plugin, we can see that what DeepSeek’s API can do is essentially what LLMs can do, the most common of which is analyzing, summarizing, and processing text content.
Thus, based on this capability, combined with some simple AI development, interesting things can emerge.
2. Bilibili Danmaku Analyzer
This was the first small tool I tried to create with AI programming. At that time, I didn’t know how to use Cursor; I just copied and pasted back and forth in Claude’s chat window.
The code was written by Claude, and the API called is DeepSeek.
I have already hosted this application on my Alibaba Cloud; here is the link, feel free to copy and paste it into your browser to try it:
This tool has the following functions:
① Get basic video information and download danmaku
② Generate danmaku distribution and word cloud charts
③ Danmaku AI analysis (this calls the DeepSeek API)
④ Danmaku & subtitle AI cross-analysis (this also calls the DeepSeek API)
Bilibili’s new videos generally have AI-generated subtitles with timestamps, but the official API interface is not provided, requiring packet capture to obtain. This operation is more convenient with a browser plugin, so I added an upload function for subtitles.
Besides DeepSeek, other AI APIs were also used here.
It can facilitate multiple AIs to discuss a single topic in multiple rounds and extend new viewpoints during the discussion.
For example, there is currently a question on Zhihu’s hot list:
I used this topic to have five AIs discuss it for five rounds.
Five AIs began speaking in order, starting from the first round,
discussing up to the fifth round.
Finally, a representative was chosen to summarize the previous five rounds of discussion.
Large language models have some ability to recognize typos, but it can be frustrating as they often miss some or excessively detect others.
So, without fine-tuning the model itself, I chose to use three different models to check the same article. By cross-verifying their results, I can improve the overall fault tolerance of the checking process.
5. Using with Feishu Multidimensional Tables
Earlier applications involved developing web applications using AI programming tools to call APIs. Are there any productivity application scenarios that do not require development or servers? Yes, the most typical example is automating with Feishu Multidimensional Tables.
For example, the table in the screenshot below:
This table’s function is to store the full text of articles published by Zhihu bloggers in the multidimensional table. After clicking the button, it calls the DeepSeek API to exclude parts of the full text content cell that do not belong to the main text and then analyzes and summarizes the author’s professional background based on the main text content.
In fact, Feishu Multidimensional Tables itself has AI capabilities, but one is that its ability is relatively weak, and the other is that there are many input limitations. For example, in this case, extracting the full information of an entire article using Feishu’s built-in AI will directly prompt that the input word count exceeds the limit, so external API calls can solve this problem.
All you need to do is set the trigger conditions in automation and send the HTTP request.
As for how to write the request URL, headers, and body.
I chose to ask AI directly, and received the correct answer provided by AI.
Finally, just write the returned values to the corresponding positions in the table.
6. AI Automation Programming Plugin
I have now used two VS Code plugins, one of which is well-known, Cline, and the other, Roo Code, formerly known as Roo Cline, is a branch of Cline. It can use one thinking model as an architect and another model to execute the code.
Although they are VS Code plugins, I recommend using Cursor for installation, as Cursor itself is also based on VS Code, and these two plugins can work seamlessly in Cursor.
Overall, these two plugins are quite similar, and even the configuration interfaces are nearly identical; you just need to fill in the API Key:
If you use Roo Code, you can create two sets of configurations on this page, using the DeekSeek-R1 model as the architect and the DeekSeek-V3 model to write the implementation code.
Using it is just like a web chat tool; you directly state your request, for example, “Help me create a calculator.” DeepSeek responds with its understanding and quickly starts generating the code.
In no time, it generates complete HTML, CSS, and JS files and prompts me to Run Command to open them.
A calculator application is created just like that, taking less than three minutes.
This is also a new paradigm for solving problems in the AI era, in my opinion.
In the past, those who mastered some search capabilities, including myself, would reap a portion of the small benefits. Because we could always judge whether there might be existing solutions for certain problems, and then search for them to find these ready-made solutions or small tools available on the internet. We would then use these existing solutions to quickly resolve new problems we encountered.
But with the advancement of AI, the optimal path to solve many problems may no longer be to find a good existing solution, but to directly use AI to generate a 100% matching solution for the present problem.
Let me give two small examples, both of which are small tools I directly generated using AI to solve problems.
This tool cleans up spaces, empty lines, garbled text, ads, Markdown tags, etc., generated by copying and pasting web text.
Many large language model responses use Markdown format. Markdown is indeed very lightweight, taking up no tokens while allowing for various formats. However, when we copy AI’s responses, the content often comes with a host of Markdown tags, which can be messy if the final usage scenario does not support Markdown.
So I directly used AI to generate a Markdown converter, where the content with Markdown tags pasted on the left will automatically convert to the corresponding rich text format on the right, allowing for easy copying and use of the converted content.
These two small tools certainly exist online. However, finding the right free and effective ones may take longer than the time spent generating them directly with AI.
In addition to the two Cursor plugins mentioned earlier, Cline and Roo Code, another recommended plugin is the Remote-SSH plugin.
This plugin allows you to connect directly to cloud servers, generating code and files directly on the server, eliminating the hassle of deploying locally written programs to the cloud and saving the trouble of configuring the development environment on your local computer. (Of course, this is only regarding extremely lightweight AI web applications; professional developers are another story.)
I recommend renting a cloud server so that you can host your generated AI applications in the cloud for easy access, just like saying goodbye to Baidu! I created my own personalized URL navigation page using AI.
A 2-core 2G server is generally sufficient, for instance, on Alibaba Cloud:
https://www.aliyun.com/minisite/goods?userCode=r18u1tal
The configuration for 68/year and 99/year is basically the same, with the difference being that the lightweight server will increase in price after the first year, while the ECS economical instance will still be 99/year for the second year renewal.
If, like me, you plan to host a long-term URL navigation page on the server, I recommend choosing the second option; if it’s mainly for testing and temporary use, then the first option will suffice.
API calls have now become so simple that you can find the API documentation, copy it, send it to AI, and then send the API Key to AI, and AI can automatically write the code to call the API.
Now, what separates ordinary people from ordinary developers may no longer be technical ability, but rather the ideas to meet needs and the motivation to try.
