MeshPilot: High Quality 3D models for AI agents in one API call

The easiest way for AI agents to generate production-ready 3D models - for games, AR, 3D printing, and design. One HTTP call, pay per use in USDC over x402.

Why I Built MeshPilot: My name is SSM and I’m a Canadian-Indonesian 15 year old, and for the last few months I’ve basically lived inside AI agents. I kept watching them do incredible freaking things. They could write full games, design systems, even plan entire experiences. It was just so cool. But the second the conversation turned to “now make the 3D stuff”, everything broke. The agents would hit a wall. Humans would hit the same wall. Suddenly you were either: Opening Blender and losing hours Hiring someone on Upwork for $100+, Or just giving up and using whatever generic assets you could find It felt ridiculous. We have models that can generate images, video, and code in seconds… but 3D was still slow, expensive, and manual. So I decided to build the thing I wished existed: an on-demand 3D printer for AI agents. I wanted something so simple that an agent could just say “I need a detailed sci-fi mech with detachable parts” and actually get it, no API keys to juggle, no complicated dashboards, That’s MeshPilot. It’s a single REST endpoint. Your agent makes one call, pays a few cents automatically in USDC on Solana (using the x402 protocol), and gets back a clean, textured, production-ready GLB. I added the features I actually needed while building with agents: Auto-rigging so characters can move Segmentation so you can ask for “just the left arm” or “just the helmet” and get separate pieces that snap back together Model safeguards: vision check before generation + automatic cleanup of floating junk so you don’t get broken meshes Easy iteration: keep refining the same model conversationally. And Animation which is still work in progress. I built this after school, teaching myself 3D pipelines, payment systems, and how to make things reliable enough that agents (and humans) could actually depend on it. It’s still in beta, and I’m learning a ton as I go. But the core idea is simple: make 3D as easy and instant as prompting for code or images.If you’re building with agents, making games, working on robotics sims, or just want to prototype ideas faster, this is for you. Try it, tell me what breaks, and let’s keep pushing what agents can actually build. — SSM Founder of MeshPilot

A pay-per-use API for 3D generation is a really clean idea. One thing I'd love to see is a way to pass reference images or style constraints alongside the prompt, so you can steer the output toward a specific art direction instead of getting whatever the model guesses. Would make it way more useful for matching an existing game's look.

@senarcyl Love this. And you nailed exactly where it was weak, so I just shipped it. You can now pass reference images as art direction: the subject comes from your prompt, but the palette, materials, shading and detail get matched from the reference. So "a health potion" + a screenshot of your game returns a potion that fits your game's look, instead of a generic one. There's also a text style knob now (e.g. "low-poly PS1, flat shaded") if you don't have a reference handy. It's live, images and style on POST /api/generate, docs at meshpilot.cc/docs. Thanks for the nudge, this made it genuinely more useful.