TypeUI
Build better UI with AI
179 followers
Build better UI with AI
179 followers
TypeUI is an agent-first product that you can use with your favorite AI tools like Claude, Codex, and Cursor to start building UI that makes sense an converts users using human-curated design skills and prompts.
Free Options
Launch Team / Built With
Viktor.comAn AI coworker that actually does the work
Promoted
Loooong time no see, PH 🙌
We are the makers of Flowbite and we had such a blast the past few years.
A lot of things have happened since then... LLMs, AI, agents, a revolution really.
After a long vacation we are back with a revenge and this time, we set out to fix UI generation using AI. It's a hell of a challenge to undertake, given the complexity and subjective essence of what UI means and how users perceive it.
First of all, we believe that good UI is the one that gets the message across, the one that is accessible for all users, and ultimately converts your users into customers.
About 3-4 months ago we launched on Reddit (https://www.reddit.com/r/ClaudeA...) a collection of markdown files (design skills as we call it) and it had quite a success. Over 1000 upvotes and many more k's of views.
A few days later Google launched Stitch and the DESIGN.md format.
Fuck, we're now competing against Google?
And then Claude Design comes along?
And then v0 launches design systems?
We have no chance...
Other than, we kept on and didn't give up.
One month ago we launched our MCP server (https://www.typeui.sh/docs/mcp-s...) that is a superset of the design skills that we previously launched (and perfected since with more granulated .md file breakdown for design system structure) and what it does is pretty simple.
It brings together all resources from TypeUI (markdown files and prompts) and lets you generate much better and cleaner UI by simply typing stuff like:
- "Build me a landing page using Atlas design system"
- "Iterate 3 more variations for the hero section"
- "Build me a dashboard with user analytics"
A lot of things happen in the background with TypeUI MCP:
- it will select the design system markdown files from your project
- it will select the best prompts available for layout generation
- it will use cleanup loops using our UI/UX fundamentals package
- it will optionally create multiple variations of sections and pages based on your settings
Ohhh, and you can also create your own brand kit (https://www.typeui.sh/create/brand-kit) which creates a BRAND.md file that you can combine with any of the design skills from TypeUI... so basically create anything in any style!
We have about 3-5k new user sessions with our MCP server... DAILY!
Anyways, back to the product.
You can also use TypeUI with autonomous agents like Hermes (https://www.typeui.sh/docs/guide...), OpenClaw (https://www.typeui.sh/docs/guide...), and Eve (https://www.typeui.sh/docs/guide...).
Because TypeUI is an agent/AI-first product, you can let them loose and automatically generate clean UI if you want your agents to build websites, dashboards, or applications so you can delegate the creative process completely.
ALSO... hear me out.
Tokens... they're getting expensive!
We see posts about Fable generating UI here and there, but nobody is talking about the fact that they are DAMN expensive. And UI generation... it requires a lot of tokens.
This is why we benchmark our previews of design skills (https://www.typeui.sh/design-skills) using Composer 2.5 from Cursor - one of the cheapest and fastest models out there. This is why a subscription to TypeUI can definitely save you hundreds if not thousands of dollars when building UI with AI.
All in all...
We reached 10k users. We have about 2.5k MRR at this point.
We are tired. We are excited. We are hungry.
We welcome all feedback so that we can improve TypeUI and help new founders and agents to build user interfaces that will connect with people, make them accessible, and keep a more clean and consistent internet.
Thanks for reading.
Love you all ❤️
Zoltan
@zoltanszogyenyi I think you created a great product here, Zoli.
And seems the launch went really well. Good luck!
@csaba_kissi thanks a lot, Csaba! 6 months of work went into it so far 😁
@zoltanszogyenyi Congrats! Go, go, go! :)
the reddit -> Stitch -> Claude Design gauntlet story is a wild launch-day flex, respect for shipping into that. one thing I'm curious about mechanically: when multiple design skill files could apply to the same component and they pull in different directions (say one favors dense information layouts, another favors generous whitespace for the same brand), does TypeUI have any way to arbitrate that, or does it just hand all the context to the model and let it figure out the tradeoff itself
@galdayan
we recommend using only one design skill from TypeUI because a design skill = a design system
if you use our whole stack (via MCP) you also get a UI/UX fundamentals skill installed automatically
@zoltanszogyenyi good tradeoff honestly, one skill = one coherent system beats stitching together partial ones. curious if the UI/UX fundamentals skill ever overrides something a more specialized skill would've called differently, or is it strictly additive on top
This is a very interesting direction, especially coming from the Flowbite team, and it is always nice to see a Hungarian maker building something ambitious here :)
AI UI generation is getting better, but a lot of outputs still feel like "nice looking screens" instead of actual product UI that converts, stays accessible, and fits a real brand. as someone building a visual product, the design skills / BRAND.md approach feels much more practical than just prompting harder. Curious how much TypeUI improves inside one project over time. does it learn from accepted UI variations and brand decisions, or does it mostly rely on the provided design skills and markdown context?
@andrasczeizel
we're not a lot of hungarians out here, but we make it count, haha
so TypeUI aims to give the best possible context for UI generation to any AI models
currently, we provide design skills (as in markdown files), prompt content (for layout generation), and via the MCP we provide variation + cleanup loop tasks to make sure that the UI stays relatively clean
now...
I actually believe that it is not TypeUI, or any AI model that should have the last say in things
I believe the user, meaning you, the builder, who should decide what you want to ship
all that TypeUI does is that it expands the amount of UI ideas the AI will generate and makes sure it stays within some parameters of accessibility, design system constraints, well documented layout options, and more
as a future feature, we are thinking of some analytics type of product inside TypeUI to also measure user interaction with your UI so we'll be able to make suggestions based on user behaviour, because at the end of the day that's what you really want to measure (think heatmaps)
thanks!
@zoltanszogyenyi Haha exactly, we have to make it count :)
That makes sense, and I really like the philosophy that TypeUI should expand the range of good UI ideas, not make the final decision for the builder. AI can generate more options, but taste and product judgment still need to stay with the person shipping it.
The future analytics/heatmap direction sounds very interesting too. If TypeUI can eventually connect UI generation with real user behavior, that could move it from "generate a nice layout" to "generate something that actually works better." Curious to see how that evolves! :))
Would love to see a way to preview the generated UI directly in the chat before exporting, maybe a quick screenshot or interactive snippet. Would make iterating with Claude or Cursor way faster since I could react to what I actually see rather than rebuilding locally each time.
@kerimova89863 interesting perspective, although normally you would very quickly see the actual UI being generated right in your localhost development server, and that's probably the closest thing you can get in terms of previews, the real thing
you can, however, actually preview the design systems from TypeUI
https://www.typeui.sh/design-skills/essential
here we used Composer 2.5 to generate these websites with a marketing, application, and e-commerce UI category
Congrats on the launch and the new journey! This seems like a perfect fit for the new wave of Vibe Coders that doesnt know much about the way a design system works under the hood; and still want a beautiful, consistent and "cheap" generated UI. I mean cheap in terms of the token usage and costs, not the end quality which is great. I've burnt 2-300$ in API credits for Claude Design just to get a bunch of components and HTML files that I needed to remake in Claude Code after that. Hopefully more and more people will use TypeUI as the go to for building their products base.
@axelut
thanks a lot, Alex!
yeah, I see a lot of trendy and flashy designs being created with AI becoming viral on social media, but those designs are not actually very practical in terms of larger user audiences or getting your message across
good UI means one that communicates and resonates with the website visitors, is accessible, and also loads fast and uses as little resources as possible, so the big 3D blobs and videos are mostly for agencies or more specialized websites
we benchmark against Composer 2.5 which is one of the cheapest and most accessible LLM model out there, using Fable or even Opus for UI generation is an overkill in my opinion
we have lots to improve on TypeUI still, but we believe this will be one of the flagship products to refer to whenever you want to design yourself or your agents a website, whether it's a dashboard, e-commerce site, or marketing your product
Congrats on the launch. The human-curated design skills part stood out — AI UI usually fails on taste rather than syntax, and curation is the missing piece. When Claude or Cursor uses TypeUI, what does the agent consume: a skill file it reads before generating, or live calls to your API?
@vollos
hey man, thanks for tuning in!
it uses skill files, markdowns, and prompts via the MCP server
we are planning out an API and CLI too now for extended usage
@zoltanszogyenyi That makes sense — skill files mean the agent treats it like docs rather than calling home every generation. A CLI on top would be handy. Good luck with the rest of the launch.
Curious how the "human-curated design skills" actually stay fresh over time, do you update the prompt library regularly or is it more of a set-and-forget approach? Also wondering if the output is locked to your own components or if it plays nicely with custom design systems like a shadcn variant.
@emirhan277217
we constantly curate the design skills, but I don't think it's just the skill files (aka. themes) that are the main feature of TypeUI, the reason is because TypeUI also gives you access to a private registry of design systems that we actually encourage you to customize via brand kits and your own colors, and share across all AI tools like Claude, Codex, Cursor
so the design skills are just examples as starting points, but the real benefit is using the MCP with all the UI features like variations, cleanup loops, prompt layouts, and having a custom registry of design systems