Say goodbye to installation, I found the web version of Cursor (with 4 new AI programming scenarios)

As a Cursor long-term subscriber,

I thought Trae's free Claude3.7 and DeepSeek-V3-0324 were good enough.

In addition to daily code development, make some small tools that you urgently need, say goodbye to annoying usage limits and advertisements, replace PPT with visual pages, etc.

Although the big models are very powerful, I still encounter a problem when I recommend them:

Is there a platform that brings together workflows such as generation, deployment, sharing, and secondary modification, so that you can skip dependencies such as Cursor and Trae that require a code environment and implement your first AI application as easily as possible?

Yes! I just found a web version of Cursor:

Snap

?  haisnap.Com

Based on my experience, this should be the answer for the new version.

In short, Snap combines the multi-model calling and online deployment capabilities of YOURWARE.Requirements GatheringandCode Generation:

The collection process will produce a to-do item, and a snap of the fingers will automatically provide more details. I can choose whether to add the implementation. During the requirements refinement and code generation process, the front-end and back-end will be output first, and then the UI components and page layout will be adjusted. If any syntax errors are found during the generation process, they will be corrected for the second time on the spot, without having to wait until the next conversation.

Let's warm up with a plate of snake.

It feels pretty good, much better than the last time I made a special structured prompt + DeepSeek V3 to get the snakes.

From the complete generated video, we can see the principle behind the good effect.

First of all, the finger snap probably uses prompts or reasoning models to collect user needs. The interaction of actively collecting user needs before generating results is currently more common in OpenAI o1's Deep Research. The general reasoning model tends to generate directly after thinking.

Secondly, from the perspective of the code generation structure, the structure generated by the snap is more complete than the single code file I manually generated with Cursor. Now the workflow of many visual web pages and stand-alone games is to ensure that the model has as few dependent environments as possible, while packaging the implemented logic into a code file and then copying it to Yourwave.

But in fact, this generation method is very picky about the model, and some models with good coding ability can easily fail in this way. The reason is that when the model learns code data, it has more hierarchical code structures, such as file A for UI, file B for interaction, file C for data storage, etc.

The generation method of snapping fingers is equivalent to being more in line with what the model has learned, effectively liberating their coding capabilities.

Let's make a visualization web page.

This is also a high-frequency problem of DeepSeek V3, Claude3.7, and Gemini2.0. In short, after recognizing the podcast audio into text, the model must rely on its own understanding ability to divide the topics in the large dialogue text and accurately generate web pages.

First of all, the page is quite complete, with podcast as the theme, and outputs the latest podcastsummary,Further reading,Smart GraphandSubscription Module, the function of visual webpage is quite comprehensive, especially when combined with variousDeep SearchThis function can turn a report of tens of thousands of words into a web page. You don’t need to worry about the file size and can just share it directly.

Let's take a look at the 3D effect animation demonstration that even V3 can't handle.

Snap your fingers straight out

Compared with the new version V3, Snap Finger also makes 3D modeling for each step. Compared with using V3 alone, the rotation animation of the object and the jump switching effect after clicking the corresponding step in the sidebar will be smoother, and more detailed text descriptions are given.

Keep going!

In addition to making small games and visual reports, SnapFinger also basically includes some small tools that require watching a lot of small ads to use.

Yesterday, to celebrate entering the top 20 of the new list, I spent almost an hour choosing the right lucky draw wheel. To be honest, the tool webpage that ordinary users see is full of advertisements, and you can't set rules, and it's not convenient to share.

I won't be sloppy today.

The prompt is quite simple. Since there is a need to collect this setting, there is no need to write a complicated prompt:Make a perfect lottery tool that supports modifying rules

This lottery tool, which was written in 5 minutes, supports random draws and weighted draws. We can customize the objects and weights to be drawn and save them as rules for next use. The final rotating animation effect is also quite beautiful. I decided to use this webpage to draw a few invitation codes for AI tools for everyone next time.

In general,

The single success rate of snapping fingers is very high. I think it should allow more users who have no experience in building code and complex prompts to try to generate the applications they want.

These applications can be cool interactive effects, small tools that are commonly used in daily life, or even replicas of your favorite games.

Currently, these tasks can be generated at the same time, with no limit on the number of times. Free users can publish 3 applications to the open source community. I have also published the above case.

Compared with AI programming-related applications such as Cursor, Manus, and Replit, Snap Finger can avoid local deployment, comes with deployment sharing, and can create templates with one click.

 Last words 

In the future, Ring Finger will also support MCP, access Github for team collaboration, as well as visual fine-tuning of the interface, modifying button logic and other secondary development functions.