Yiwei Ho posted every slide at the Rayboba event was built with open-slide. Not a demo. Not a mockup.
Decks for a community event hosted by Raycast. And the slides looked clean.
This matters because slide tools built for agents have become their own category this year. There is AI presentation generators everywhere. Type your topic, get a PDF.
They all promise the same thing. And they all deliver the same disappointment.
Bullet points on a gradient background. A logo in the corner. Something that looks like a presentation but feels like a template you have seen a thousand times.
open-slide does not do that. And people noticed.
The GitHub repo hit around 4,000 stars in about a week. The creator posted a video of Cursor building a deck in under a minute.
212 likes, 16,000 views. Replies poured in.
"So cool!" "This is awesome." "Yo lfg."
But also real questions. "Is it PPTX compatible?" "Can I export to PowerPoint?"
The excitement was loud. So was the hesitation.
I spent last weekend reading every reply. Wanted to understand what made this different from the other AI slide tools I have tried. And I found something I did not expect.
It is not a slide generator. It is a slide framework.
What open-slide actually is
open-slide is a React-first framework. Every slide is a React component on a 1920x1080 canvas. No templates. No drag and drop.
No limited layouts.
You describe your deck in natural language. Your coding agent writes the React. open-slide handles the canvas, navigation, hot reload, and present mode.
You start with one command.
npx @open-slide/cli init my-deck
That scaffolds a workspace. Then you use /create-slide with your agent.
The skill asks four questions. Topic and aesthetic. Page count.
Text density. Motion or static.
Based on your answers, it plans the structure and writes the pages as React components.
The first time I read this, I thought it sounded like extra steps. Why not just use a prompt and get a PDF?
But here is the difference. A PDF is dead. A React component is alive.
You can edit it, inspect it, change anything.
I have tried the AI presentation generators. NotebookLM gave me the closest thing to a real deck.
But every edit meant rerunning the whole prompt and hoping for the best. That is not editing. That is gambling.
open-slide works differently. The output is not a black box. It is a file in your project.
You can open it in your editor. Fix a typo. Swap a color.
The agent owns the creation. You own the file.
The inspect and comment loop
This is the part that got me.
In the dev server, you can click any element and attach a comment. "Make this red." "Change the headline." "Shrink this font."
Those comments get saved as markers in the source code. Then you run /apply-comments and the agent edits the files. It clears the markers when done.
The loop is simple. Present. Click to comment. Apply.
Repeat.
No back and forth about what "section 3" means. No regenerating the whole deck for one color change. You point at the thing. The agent changes the thing.
What happened on X
The reactions split into two groups.
One group was excited. Vinay Juneja called it "damn fast and soo crzy shipping man." Alexi said "Someone has been on fire lately!"
The other group asked practical questions.
fil asked "is it pptx compatible?" Elias Lumer asked "Can u export these slides into an edit-able PowerPoint?"
These are not complaints. They are constraints. Most people do not deliver from a terminal. They need a file to share. A format their team can open.
But one reply stood out. vinaykumar wrote: "Decks are a good stress test because layout exposes bad agent state fast. Harder to fake than a clean text diff."
This is the honest take. open-slide tests how well your agent handles layout. And layout is where most agents fall apart.
There is another angle here. open-slide works with any agent. Claude Code. Codex. Cursor. Gemini CLI.
The framework does not care which one you use. Each agent gets the same /create-slide skill and the same canvas rules. The quality difference between agents shows up in the deck.
That is the stress test.
One command exports the whole deck as a static HTML site or a print-ready PDF. No server needed. One click deploy to Vercel, Cloudflare Pages, Netlify.
This is the part I think gets overlooked. You are not locked into anything. The output is plain static files.
Here is what it does:
- Agent-native authoring with built-in skills
- In-browser inspector with comment editing
- Assets manager with svgl logo search
- Present mode with speaker notes and timer
- Export to static HTML and PDF
- Slide manager with folders and drag and drop
Here is what it does not do yet:
- Export to PPTX
- Drag and drop editing for non developers
- Work without an agent
The thing I keep thinking about
I spent an embarrassing amount of time searching for "open-slide" and getting results for OpenSlide. You know, the medical image viewer. Thousands of pathology slide scans.
It has been around for years.
This happens every time I name a project. You search for something obvious and find three other things with the same name.
I once named a side project "Tracer" and found twelve SaaS products using the same name. Maddening.
open-slide is stuck with this for now. The SEO overlap is real.
But maybe that changes if the project keeps growing. The star history graph is vertical. That kind of growth helps.
I also wonder about the name itself. "open-slide" sounds like it describes what it is. Open slide framework.
But it also sounds like open-source slide. And that is accurate. It is MIT licensed.
You can fork it. Modify it. Build your own workflows on top of it.
That matters more to me than the name confusion.
But I still wish naming things was easier.
Honest talk about who should use this
Most people do not need this.
If your presentations have eight slides and bullet points, use Google Slides. It is faster. Your team can edit it.
No one needs React for a status update.
If you work in a corporate environment where everything must be a .pptx file, this tool is not for you yet. The export question came up multiple times in replies.
That is a gap.
open-slide makes sense when your deck has visual density. Charts. Custom layouts. Animations.
When you want each slide to feel intentional instead of templated. And when you already have an agent setup ready.
Do not pick up a React framework for slides unless you have the agent workflow figured out. The tool is not the hard part. The agent workflow is.
A final thought
I keep coming back to that one reply. "Decks are a good stress test because layout exposes bad agent state fast."
This is the signal. Not the 4,000 stars. Not the landing page.
open-slide reveals where agent coding is right now. It can handle the hard part. Layout.
Visual design. Complex positioning.
The part that usually gives agents away as robots.
But it also reveals the gap. People want to share these files. People want to collaborate.
People want a format that works outside the dev environment.
I think about the Rayboba event. Real slides. Real audience.
Real presentation. That is more than most "AI slide tools" have achieved.
What is the point of a perfect slide if you cannot email it to your boss?