Idea: AI chat components for HTMX + daisyUI apps

[hqdrt]

Hi all,

First off, thanks for daisyUI. Most of my recent projects are built on it, and it's a big part of why I can stick with server-rendered HTML instead of pulling in a JS framework.

The pain

Over the last years I've built a handful of AI agent apps: chat UIs where the agent streams its answer, thinks, calls tools and cites sources. My stack is Django with HTMX, Alpine.js, Tailwind and daisyUI, and a Python agent SDK (went through them all: LangChain, custom ones, and now Pydantic AI) behind it.

I end up rebuilding the same pieces from scratch:

• a prompt input with attachments, a model picker and a stop button
• streamed markdown that doesn't flicker or jump around
• collapsible "thinking" blocks
• tool calls (running, done, failed, or waiting for the user)
• sources and inline citations
• message actions, follow-up suggestions, a context meter

None of it is hard on its own, but getting the details right (streaming, scrolling, all the states, accessibility) eats a lot of time, and I keep solving the same edge cases on every project.

Meanwhile in React land

React folks have this pretty much solved. The Vercel AI SDK streams from the model to the browser, shadcn/ui gives them components they copy in and own, and Vercel's AI Elements (built on shadcn/ui, made for the AI SDK) ties it together with ready-made chat components. You can get a polished agent chat running in an afternoon.

I haven't found anything like that for server-rendered apps with HTMX and daisyUI.

So, two questions

1. Has anyone else run into this building AI or chat UIs with daisyUI? What did you end up doing?
2. Would people be interested if I took a first stab at it? I'm picturing a small set of copy-paste AI chat components built from daisyUI classes, plus some notes on streaming them from the server with HTMX.

I'd also love to hear from the maintainers on where something like this should live, if it's useful at all.

Happy to share more about what I've tried so far. Thanks for reading!

[saadeghi]

daisyUI as a CSS library can cover the visual parts (chat bubbles, textarea, etc)

In the next versions of daisyUI we can add the visual for a "composer" component which would be a combination of a text area, buttons, dropdowns etc.

The parts that are about the functionality, streaming data, or interacting with the server are out of daisyUI domain.

Please let me know if I can help with anything.

[hqdrt]

The parts that are about the functionality, streaming data, or interacting with the server are out of daisyUI domain.

Totally agree, the rest would live separately built on top of daisyUI, not in daisyUI itself.

Let me compile what I have in mind with an actual example and we can discuss and review what could be covered in daisyUI.

[hqdrt]

Here a draft so we can discuss it further: https://htmx-ai-composer.pages.dev/

Open questions:

1. Name. composer or something AI specific?
2. Class or attribute? Busy is composer-busy; the form also has aria-busy="true" while busy. Style that instead, or both?
3. Busy animation. Sweeping highlight via @property, conic-gradient and a mask. Too much? A static border or a pulse would work too.
4. Toolbar. Worth a class, or just a flex row with utilities?
5. Sizes and colors. No composer-sm/composer-lg or color variants yet. Wanted?

[pdanpdan]

I don't see the point of doing something specific for this when am the components used are basic existing ones.

It's mainly this: https://play.tailwindcss.com/02y2Slc84F with bells and whistles based on your specific requirements

[hqdrt]

Thanks for taking a look, and @pdanpdan good call. I threw together a quick side-by-side: https://htmx-ai-composer.pages.dev/compare/

Left is the composer as a new primitive. Right is the same thing built only from what daisyUI already has: textarea for the frame, aura for the busy state. You can flip between busy / drag-over / disabled, and toggle dark mode.

You're right: everything's already there, and the two look basically the same. So the question is more about whether daisyUI wants a ready-made, opinionated component for it.

Pros

• It's daisyUI's own idea: about 4 short lines with a composer class, against about 10 lines of utilities on top of textarea (see the page). Replacing a pile of Tailwind classes with a few daisyUI ones is what daisyUI is about.
• People building AI chats who are torn between daisyUI and something else might take it as one more reason to go with daisyUI.
• shadcn/ui has a generic component for this exact shape: Input Group, a textarea with addons/buttons above or below. daisyUI's input already covers the single-line case (wrapper with icons/kbd inside), but I didn't find anything for a textarea with a toolbar underneath.

Cons

• Nothing's actually missing (see the page), so it's more to maintain for something that's mostly convenience.
• The AI-specific bits don't belong in daisyUI. shadcn draws the same line: their Prompt Input is a recipe in AI Elements, built from Input Group and other core pieces, not its own component.

What do you think?