2
1 Comment

I spent a whole weekend fighting Chinese payment gateways just to test DeepSeek. Then I snapped and built a bridge.

It was 2 AM on a Sunday. I was staring at a Chinese SMS verification screen that kept timing out, holding my credit card, practically begging a platform to just take my money.

All I wanted to do was test an API.

For the past few months, the benchmarks for models like DeepSeek-v3, Qwen 2.5, and Wan 2.1 have been all over my X and Reddit feeds. They are insanely capable, incredibly cheap, and I desperately needed to plug them into an AI workflow I was hacking together.

I figured it would be a standard 10-minute job: sign up, grab the key, swap the endpoint.

Instead, I descended into what I now call "The Cross-Border Developer Hell."

First, I hit the Identity Wall.
To even look at an API key, I needed a +86 mainland Chinese phone number. I spent three hours hunting down burner number services online. None of them worked. I eventually had to wake up a friend halfway across the world just to borrow their phone for a 6-digit code.

Then came the Payment Wall.
I finally got in, successfully generated a key, and went to top up my balance. I pulled out my Visa. Rejected. I tried Stripe. Not supported. Crypto? Forget about it. The only options staring back at me were WeChat Pay and Alipay, both of which require a local bank account to function.

Finally, the SDK Nightmare.
By some miracle, I bypassed the payment issue. I went to integrate the top-tier video and image models, only to realize their API documentation was an entirely different beast. Nothing was standardized. Trying to jam them into standard OpenAI wrappers broke my entire codebase.

By Sunday morning, I hadn't written a single line of logic for my actual product. I was just fighting infrastructure.

I sat back and realized something: I work in the digital space, and I was still pulling my hair out. If I was struggling this much, a developer sitting in San Francisco or London had absolutely zero chance of accessing these models. It was a massive, untapped information gap. The models were world-class, but the administrative friction was keeping the global dev community locked out.

So, I trashed my weekend plans. I stopped building my app and spent the next few weeks building a pure, silent middle-layer to swallow all that friction.

What I Built (And How It Works Under the Hood)
I packaged everything into a unified gateway designed to be completely frictionless for international stacks. Here is exactly what it handles:

100% OpenAI-Compatible Routing: You don't need to learn a single new SDK. You literally just change the base_url and swap in your API key. It works out of the box with LangChain, LlamaIndex, or raw fetch requests.

I didn't stop at text. The gateway maps text, image, voice, and video models under one roof. You can call DeepSeek-v3, Qwen 2.5, Kimi, Doubao, Wan 2.1 (Video), Tongyi (Image), and MiniMax/Fish Audio (Voice) natively.

Global Payments Solved: I integrated international credit cards and USDT directly into the billing dashboard. No WeChat, no Alipay, no identity verification required.

Transparent pricing: I set it up to just charge a flat 4.5% service fee over the raw domestic API costs. No hidden margins, no marked-up token rates.Watching the first successful generation come back to my local terminal in seconds, knowing the request had cleanly bypassed all that administrative nonsense... that was a better feeling than launching any feature I’ve built this year.

I originally built this just to scratch my own itch, but the infrastructure is solid, highly optimized for low latency, and running smoothly. I know there are other indie devs out there stuck behind the exact same walls, settling for more expensive Western models simply because the friction to switch is too high.

I'm keeping this unlisted for now while I gather early data, but I want to invite a few people who are actively building AI products to test it out. If you want to bypass the VPNs, the phone numbers, and the weird payment gateways, drop a comment or DM me. I’ll send you the link and toss10000 free tokens into your account so you can test it on your own stack.

on June 15, 2026
  1. 1

    What makes this interesting to me isn't the infrastructure problem itself.

    It's the conclusion that gets drawn from solving it.

    The same experience could support several very different businesses, and early traction can make one interpretation feel more obvious than it actually is.

    That's the part I'd spend the most time pressure-testing before getting too attached to a particular direction.

Trending on Indie Hackers
6 weeks solo, 2 rejections, finally live but nobody told me marketing would be this hard User Avatar 138 comments I spent more time setting up cold email than actually selling. Here is what fixed it. User Avatar 33 comments I just wanted to taste AI coding tools. A week passed. User Avatar 22 comments I built a PDF API because every team I know has a haunted corner of their codebase they never want to open User Avatar 16 comments Building LinkCover – Day 3: Payment is live. No more building, time to sell. User Avatar 15 comments I Was Bypassing Every App Blocker, So I Built One That Fights Back User Avatar 12 comments