I built a web-based vector editor from scratch and integrated an AI Agent. Need just ONE beta tester!
Hi Indie Hackers,
I’ve been coding solo behind closed doors for months, and I finally built Rayzia.com — a lightweight, web-based vector editor.
But here’s the feature I’m most excited about: I integrated an AI Agent powered by Claude. Instead of just exporting flat, messy SVGs, this agent actually controls the editor's tools (like createShape, addGradient, setPaint) live on your canvas via MCP.
Once the AI finishes drawing the paths, you can instantly grab any node or layer and edit it manually right in your browser.
Only 4 days left until the free Beta launch on Product Hunt!
I'm limiting the first batch of testers to keep server costs manageable. Drop a comment below if you want to grab an early invite link!
Since I’m a solo developer, getting even one person to test this feels like climbing a mountain. I would be incredibly grateful if anyone here could try it out and give me raw, honest feedback. Does it work smoothly? Is the AI agent actually useful for your workflow?
Check it out here: https://rayzia.com/vector
Thank you so much!
The beta test I'd run is one round trip: import an external SVG, ask the agent for a structural edit, tweak two nodes manually, then prompt again. If layer names, paths, and manual changes survive that loop, the agent is collaborating rather than just generating into a canvas.
That’s the ultimate test of a true collaborative AI! You nailed exactly what makes building Rayzia so challenging yet exciting. Preserving the user's manual edits (paths, layer structures, nodes) while letting the AI iterate on top is indeed the core goal. I’d love for you to put Rayzia through this exact loop during our Beta! Would love to hear how it holds up.
I cannot promise a beta test here, but you can make the loop measurable: hash the scene graph after each manual edit, then assert untouched objects keep their IDs, hierarchy, and geometry after the next prompt. A clean-looking canvas can still hide destructive rewrites.
Got it. That’s a brilliant way to track it programmatically. Thanks for the awesome suggestion, really appreciate the insights!
Needing just one honest beta tester is a smart way to lower the ask, way easier to get a real yes than pitching a full beta commitment. The top comment's reframe is the interesting part though, tracking what people manually fix after the AI finishes tells you more about the real gap than watching what the AI generates. Once you have that beta tester, what's the first thing you're hoping to learn from watching them use it?
Spot on! Watching where users manually intervene is where the real gold is. For my first beta tester, the number one thing I want to learn is where the 'creative friction' happens. I want to see exactly at what point they feel the AI didn't understand their design intent and had to manually correct it. That gap is where we will find the most valuable features to build next.
I quickly played with the editor for 30 seconds, feels similar to Inkscape. One feature that'd be an a must for someone like me is the ability to copy and paste in external SVG graphics. I work with both SVG code in VS Code and vectors inside Illustrator frequently — so being able to copy in paste the graphics around is core to my workflow ... including into tools like an optimizer or SVG to CSS convertor.
Thanks for the awesome feedback, Visiwig! I can definitely make that happen for you. Copy-pasting external SVG graphics/code directly onto the canvas is a fantastic addition for a pro workflow. I'll get to work on this right away. Stay tuned!
The interesting opportunity isn't putting an AI agent inside a vector editor—it's making design iterative instead of generative. I'd keep validating which edits people still make after the AI finishes. Those moments will tell you where the real product advantage is.
Thanks for the awesome feedback, Aryan! You totally nailed it.
That is exactly the core thesis behind Rayzia. Most generative tools today are "one-shot" and leave you with a flat, locked image. I want the AI to act as a collaborative partner on the canvas—where you can prompt it, let it draw, then manually tweak it, and then prompt it again to build on top of your manual edits.
Tracking what users edit manually after the AI finishes is a brilliant suggestion. I will definitely keep a close eye on those patterns during this Beta phase. Thanks again!
I'm glad it resonated.
Reading your reply gave me one thought about what those manual edits might reveal about the product direction. I'd rather explain it in the context of Rayzia than try to reduce it to a few comments.
If you're interested, what's the best email to reach you on?
I'd love to hear your thoughts on this, Aryan! That sounds incredibly valuable.
You can reach me directly at [email protected].
Looking forward to connecting and diving deeper into this!
Thanks! I’ve just sent it over.
Looking forward to hearing your thoughts whenever you have a chance.
Hey Aryan, it seems my custom domain email is having a slight delay and hasn't shown up yet.
Could you please resend it to my backup email at [email protected] instead?
Sorry for the trouble, and thanks a lot!
Thanks for letting me know.
I've resent it to [email protected]. Hopefully it comes through this time.
Looking forward to hearing your thoughts whenever you get a chance to read it.
I'll test it. Do the layers come out clean enough to actually edit after, or is it nice-looking paths that are hell to touch? That's the whole reason to drive the tools instead of dumping an SVG.
Spot on! That’s the exact frustration I had with other tools, and why I took this approach.
Since the agent literally drives the same underlying tools (like drawing shapes and adding gradients step-by-step), it behaves just as if a human designer built it. The layers are generated as native canvas objects, meaning they are fully clean, grouped, and easy to grab, scale, or edit manually.
No messy, autogenerated SVG spaghetti code here.
I'd love for you to put it to the test and break it! Sending you the direct Beta link right away. Let me know what you think of the layer structure.
Nice, got it. Will try to break it this week and tell you if the layers hold up.
Awesome! Can't wait to hear your honest feedback.
Go ahead and try to break it as much as you can—if it breaks, it just means I have a great opportunity to make it even stronger.
Have fun testing, and let me know whatever you find!