Show HN: Rapidpages – OSS alternative to vercel's v0 | Hacker News Reader

Show HN: Rapidpages – OSS alternative to vercel's v0

github.com

216 pointsmuratsu50 comments

Hey everyone,

Really excited to share what I've been working on. Rapidpages is a prompt-first online IDE, think midjourney for front-end developers. I've been working on this for a while and it's great to see some interest from companies like Vercel in this space.

All you need for self-hosting is an OpenAI key and a GitHub oauth app. Simply clone the repo and play with it. It's also available on the cloud at www.rapidpages.io

Please give it a try and let me know if you have any feedback, and if you like what I'm doing with Rapidpages, please give it a star on GitHub.

Thanks!

Open articleView on HN

Great work! I think the barrier to entry here is going to gradually become smaller and smaller (as models become better and we see more multi-modal models) and the big differentiator here is that you will have lots of open source developers building functionality that is relevant to them (vs keeping this closed-source like Vercel).

I am curious how much of the functionality in this area is "write code that does scaffolding and prompt engineering" vs "finetuning and model-level improvements".

I think this kills the CMS market—-companies like Contentful & Sanity that charge $400/mo+ just for a team to be able to edit their own website. There’s been a lot of money made (and web stack complexity added) because of how much domain-specific skill is required to created and edit html content.

The only missing piece is direct framework integration (Next, Astro, etc) but at the rate of development I've seen for these projects we'll have that by next week. Then you have an intuitive site builder that operates directly on the page code just like another dev on the team.

One thing Vercel got really right is being able to click on any element in the page for it to become the context of your prompt. That way, you don't have to type "change the top icon inside the third pricing box to red". Also you'll get faster responses because you don't have to feed the AI the entire page / component each time.

Hard disagree, this ain't gonna kill CMSs, what are you talking about?! Content management goes beyond just a website, especially in the headless cms world. Also those products can be as low as free.

Been involved with a lot of Contentful projects large and small that only existed to publish content to web properties. I guess it kills that use case.

Depends imo. If it's purely just a few static pages then yeah I can see this happening but as soon as you have anything like a blog you're going to want some sort of content management system surely to manage it?

What I see this more doing is eliminating the need for someone to code the glue between the CMS and the website -- you can just ask the AI to add code to get your blogs from Contentful

There's one huge not-immediately-obvious element there: Vercel bandwidth is expensive. My company switched our CMS content over to Prismic in part because serving all media content directly from them was significantly cheaper than doing so through Vercel, even at a fairly low volume.

Vercel/Guillermo Rauch are really onto something

It's established that cloud is paying a premium for not having to hire: but they've gone a step further and started testing the limits of how much you can overcharge on top of cloud pricing for good DX

Clerk.dev (another investment of his often pitched along side Vercel) follows a similar model, and almost without fail if I see a Vercel sample pushing a SaaS, Rauch is usually listed as an investor

I wonder how it'll actually work out for them though, their revenues don't seem to justify their valuation, especially since the Zero Interest Rate Phenomenon is now over. For example, in 2022 they only hit $25 million on a $2.5 billion valuation [0].

[0] https://getlatka.com/companies/vercel

> Clerk.dev (another investment of his often pitched along side Vercel) follows a similar model

Not when there’s auth0 / Okta charging 5x more. Completely different.

Clerk.dev charges like Okta and then gets pushed to people who should be using something like Cognito which costs ~1/100th as much.

If you're not building some $100/per/month/seat enterprise tool there's no way you should be reaching for something like Clerk.dev, but obviously that'd be a constrained market.

So now it's pushed as a general bandaid for Next.js having a poor auth story, complete with the "official authentication partner" badge (but of course, no mention of the investment)

I mean Guillermo invests on a lot of Products, so do so other vercel team members, many of those end up being FOSS or Vercel friendly, hardly anything wrong there.

Pumping your own bags is legal

You're right. I've noticed this pattern, too.

Yeah - this space is moving really fast. Codesandbox (https://codesandbox.io/blog/meet-boxy-ai-coding-assistant) released the click on any element functionality back in May.

Funny because my original idea was to build a browser extension to modify existing pages with "click on any element and prompt" type of workflow. That approach also has certain limitations/gotchas too. eg if you click on a list element and ask it to push it down the list, it won't work. It'll fail as expected but it may be counterintuitive for someone playing with it for the first time. Having said that, I still believe that there's value in this UX and will most likely implement it soon.

100%!! You need one model to understand the initial prompt and figure out a development plan. Once the plan is created, the subtasks can be delegated to specialized smaller models (I think OSS will play a big role in this!).

have you explored using something more lightweight than react ? for example, wouldn't it be easier to generate vanilla html with tailwindcss ?

Sounds promising but website seems to require a login which shouldn't be required for examples.

Thanks for the advice! I'm working on building a gallery for people to be inspired next.

In case if anyone wants to see what things may look like after a few simple prompts: https://www.rapidpages.io/r/clmuv2p2r0003l808x4pzdpe6

You should just let people try it a few times. It seems a crime to get someone to your website then they click away because they can’t try it.

The hosted version doesn't seem to work at all. Any time I enter a prompt it just brings me back to login.

Ah sorry if it wasn't clear, you need to login to be able to prompt. I need accounts to prevent openai abuse.

I did log in. It still just keeps bringing me back to the login page when I enter a prompt.

That's odd! Do you have anything installed that would prevent from storing the session?

Also happy to debug further if you don't mind jumping over to discord: https://discord.gg/W6jYq46Frd

This was happening to me until I started with a much shorter prompt at first. Then once you get to the editing view, you can use longer prompts.

+1 do loop of login

Had it too, but worked with the the OP on discord. It's fixed for me now.

This is great, I'd love to see an explainer in the readme of how it works with OpenAI.

Pretty self explanatory in this section of the code base.

https://github.com/rapidpages/rapidpages/blob/c5f9eacbaba93c...

Same problem here as experienced by others.

When I try to use it it asks me to sign in with GitHub, but after having signed in with GitHub and I try to use it again, it still asks me to sign in again.

The hamburger menu on your page shows that I am logged in, displaying my GitHub username and my email address.

The prompt I used was not long either.

Do you mind refreshing and giving it another shot I've just pushed a quick fix

Thank you, now it works :)

I think the future is about to have a fork: dumb content and smart content. dumb content will be all the ai created articles, websites, music etc. and the actual creative people will move to creating much more intelligent works.

Not sure I understand the rationale behind this comment in the context of OP. How would you classify RapidPages on your measure: smart or dumb?

It will be used to create dumb sites which all look the same.

There's no shame in websites looking similar. Not all websites must have a unique design and creative layouts. Some just want to make it about their unique content, and making it look like others' helps users navigate the familiar. Most cars, dishwashers and phones look the same, while only little small changes differenciate them from competitors. Not everyone should be the Apple or Tesla of their field. Tools like this make the deliverability of content faster, in familiar layouts.

Ok this is awesome.

First ux is pretty shitty. Taking time to create a prompt, getting redirected to login, login successfully, then lose the prompt. Surely you can preserve the prompt with a cookie or even in the url.

But the product is cool after that

I can't get styling to work nicely. E.g. "make it dark mode" or "make the background dark and text light"

How difficult would it be to feed in design system components to the AI? Design system building blocks could serve to improve the quality of outputs vs straight CSS/Tailwind

Are there any open source tools that do something like that (not nessaarily with design / UI)? I'm wondering how you would 'teach' the AI about your system. I guess it would need fine tuning as the prompt would be too big otherwise.

afaik v0 uses ui.shadcn.com which uses radix ui underneath

This is a good idea. I think this is worth experimenting with. Are you familiar with any repos with good design systems? I'll take a look when I've time.

If you're interested in collaborating, I've been thinking of a feature like this for our platform (https://tamagui.dev) to integrate with the studio we're launching soon, send me an email or DM on Discord.

This would be amazing

Basic design system https://github.com/versoly/versoly-ui also not React.

Huge market outside of React for developers who just want a good looking UI.

Material UI / MUI is really popular. Fluent UI is another.

Am wondering about the process it would require to build such thing. Maybe cut out in smaller component and then prompt from this smaller component image to code?

Congrats on the launch Murat. What types of websites is it most suitable for in its current form?

Thanks!! So far, I have most success with more structured designs. B2B SaaS, simple shopping pages, and forms work the best. It’s really weak with graphic heavy designs (eg Apple).

pretty cool to include xkcd for the loading screen :)

Type a prompt and then get asked to log in. I see this as a dark pattern because if I don't want to log in, I've put data and time into your site and not getting results. If you want to keep this workflow, I'd suggest maybe letting people try it once without logging in?

Curious if you measure this and how much drop off there is between fill out a valid prompt and not logging in to see the results.

This is amazing! Have been looking at this. First steps to get prompt to Ui

Reply on news.ycombinator.com