Lovable vs Cursor: pros & cons

Discussion
1

Hello,

Any Lovable users here? Could you share pros and cons of Lovable vs Cursor?

1 Like
2

I am!

Pros of Loveable:

  • Direct preview of the website with mobile mode and select elements directly and easier reference

  • Fix error button (free of charge)

  • No need to know the programmatical terms to let the AI know what you are asking for (but not always)

  • The chat mode and coding mode are separated, which means you don’t have to tell the AI when to not code

  • Precise vision with the colors and design, layout when given image as reference

  • Able to use uploaded images directly (previously you had to go through GitHub, still have to go through GitHub for audio and videos)

  • Native Supabase integration (simplified backend)

  • Can host directly (Not fully custom domain, the domain ends with loveable)

  • Works for all devices, low-end computers friendly (even on mobile! But not recommended)

Cons of Loveable:

  • Expensive (not budget friendly, 5 uses daily, 250 uses monthly minimum), free version is very limited (can only use it for 5 prompts per day)

  • Experimental/beta/alpha features are only for $200 plan users

  • Projects are private only for paid users, free users have their projects public which can be seen by everyone even guests who don’t have an account

  • AI can get into the loop if you don’t know what you are doing and what is coding

  • Supabase can be very confusing and limiting for large projects (e.x. social media, cloud services, etc… even with pro plan on Supabase, you still limited for very limiting amount of storage, self hosted version is not natively supported which can be even more complex)

  • Does not crawl the web nor using the web generally, does not use docs easily

  • Hosting the website using fully custom domain and provider can be difficult when not guided properly, especially for devs

  • Does not code actual apps, only web apps/websites

  • Source code cannot be viewed or downloaded directly, must use GitHub which can be somewhat difficult to non tech users

  • Can only code React.js and only TypeScript, JavaScript has been removed

  • Cannot customize the theme of the chat

  • It’s web based only which can be performance issue

  • Lacks of documented syntax, only the AI knows which can be clueless when debugging

Pros of Cursor:

  • Very tweakable AI through prompts or cursorrules

  • Various models (can bring own API key to use more models from different providers other than just OpenAI but can be a bit pricey, not as pricey as Cline and Aider though)

  • Can code anything, able to use various frameworks and languages, websites, web apps, actual apps, mobile apps, especially when using @web to crawl and search the web and @docs for the desired language as it can crawl and index docs

  • Very cheap and budget friendly ($20 a month for 500 fast use of various models, then unlimited slow use of various models, intelligence does not change, only speed)

  • Can do automation when using agent mode in composer like running commands and editing multiple files at once, automatically debugs using linter

  • Can use chat mode for plans or composer normal mode or composer agent mode for all-in-one solution

  • Can deeply read and analyze everything, including the terminal

  • It is based on Visual Studio Code which is very friendly user and familiar to devs as it has a lot of useful extensions

  • The source code is direct and can be directly manually edited, no GitHub required

  • Direct terminal access whenever its PowerShell or the regular command line, which allows you to always have up-to-date packages by running the command “npm i the name of the package” for example

  • Has easy to access and use Notepads feature for extra knowledge or rule(s)

  • Experimental/Beta/Alpha features are always available for new and old users, free and pro users

  • Has pre-indexed docs

  • Easy simple drag and drop direct upload of various files as it can be directly modified through your operating system directory (but needs to be referenced properly)

  • Customizable IDE along with the chat by using themes and icons from marketplace

  • It’s native IDE app which is great for performance

  • Well documented syntax which helps the AI to know where and what to debug

  • Cursor Tabs is very fast prediction mode and very fast corrector

  • Apply model is very fast

Cons of Cursor:

  • Limited Window Context (10-20K, but with all honesty it does not really impact the performance of the AI and can still perform well)

  • AI can go through a loop when not prompted well as you need to use some programmatical terms (using XML tags sometimes helps but not always, using natural English still works like how it works with Loveable)

  • A bit older version of VSCode (1.93, but still performs well)

  • Apply model sometimes messes up and deletes half the code or comments “the rest of the code remains the same”

  • Can not handle too many lines (this is more of a LLM limitation, which means you have to refactor the files)

  • Not direct preview for websites (needs to open a browser)

  • Interface can run through visual bugs sometimes

  • Enabling Shadow Workspace mode can consume too much RAM

  • Can be heavy for low-end computers (this is more of VSCode issue)

  • The AI can hallucinate and mess up or forget some of the context including the path of the project and start working out of the project (to solve this is to create new chat/composer, and perhaps you have to give the main/basic context of your codebase otherwise it will not know what to do)

Conclusion, in my opinion as developer, I say Cursor is great choice for all-in-one solution while Loveable is just for small web projects

6 Likes
3

Really great answer! Thanks a ton!

1 Like
4

You very welcome!

5

Your comparison is very comprehensive, very good, thank you.

1 Like
6

Here’s a little secret for the folks who don’t know.
I don’t know lovable’s story but I can say their product felt clunky compared to bolt.new by Stackblitz. Given the use case of the product, I just wasn’t in the same place with the app as I would be with v0 and bolt.new.

Here’s the dope part though:
bolt.new is open sourced by Stackblitz and there’s an active community that’s been working on bolt.diy for some time now.

Best part is you can use Gemini or deepseek or whatever you want on it and you can update it however you want including system prompt since it’s open source.

It’s a little bit less polished than the production app but they are easy to work around.

That being said
Nothing comes close to Cursor agent at this moment IMO. As someone who’s prompted coming up on 8 billion tokens or so I think its better to use Cursor.

Lots of people putting in good features in there to bring it to feature parity, but there’s also a LOT that Bolt.new doesn’t have.

7

So glad to hear that! And you welcome!

8

To clear things:

The problem with bolt is that it is pricey, using your own API key always been pricey, 200K context window is pricey, even for small tasks its very pricey and using web version instead of self-hosting is very limited and not budget friendly, the results are not as good and precise as Cursor, even with being descriptive and using prompt enhancer that made by bolt is still not as good as Cursor

Now for v0, the problem with it is its very very limited for $20 which can be non budget friendly to some users, especially the very active ones and non developers, v0 is actually worse than Cursor because the precision is not that good and needs multiple shots before achieving desired result, the only good difference is that both v0 and bolt has select elements and direct preview

9

Could you explain why it is more expensive when it is web-based like Bolt and V0 than when it is self-hosted like Cursor? In both ways, the prompts are sent online to OpenAI (or Claude), right?

10

Yep I agree with all of what you said. Hence the open sourced variant where you can use Gemini which is really cheap and/or free.

2.0 flash exp or exp 1206 are great with bolt

11

Self-hosting bolt is more expensive than subscribing to their plan as using your own API key can be very pricey, same for Cursor using own API key is expensive (excluding Gemini and DeepSeek which are actually not as good as they claim, I tested them personally, they are cheap but not good, a lot of hallucinations that could cost you like how 3.5 Sonnet does)

But what I mean by both of web based Bolt and v0 are expensive plan because you don’t really get all the benefits, you have to wait for it to recharge after specific amount of usage, unlike Cursor which has unlimited usage, not to mention the performance of Cursor is being better than these two especially when using model Claude 3.5 Sonnet, therefore, that is not very budget friendly and efficient, yes, they do have bigger context window, but are they performing better than Cursor? No perhaps, so what’s the point of big context window if the output is bad

If ever, using GitHub Copilot is good alternative if anything, or v0 if you just want to have very basic components which Cursor itself can do that

12

That is true! Gemini can be free and cheap, but the performance is not as good as it seems, you’ll have to overwork it to output the desired result which can end up pricey

13

i see comparing deepseek v3 over claude 3.5 sonnet is true?

14

Like if DeepSeek v3 is better than Claude 3.5 Sonnet? Not at all! I tried it myself and its equal to ChatGPT 4o and mix of Gemini, it is faster and way cheaper, that’s true but is it better? No, you need to work hard with the prompts to finally achieve the desired result (approximate 30-80 prompt), its placebo effect when they say its better than Claude 3.5 Sonnet because its freshly new and the hype is high so they will believe the benchmarks instead of real world use cases, unless they mean the price and speed then yes, it is cheaper and faster

15

many websites measure deepseek v3 more than claude 3.5 sonnet

16

Yes! Just like how they did with ChatGPT o1 when it first came out, don’t believe the measures from these websites, believe the real world use cases and your own experience

1 Like
17

you are right

18

What is the main difference between Normal and Agent mode? I understand Agent debugs but Normal does as well when pasting in console errors.

20

Normal mode is our basic composer, being able to write code changes using the context your provide it.

Agent mode can do more on its own. Firstly, it can view your codebase structure and read specific files on its own. Also, it has additional abilities like the option to run terminal commands for you, as well as the usual composer features!