3
2 Comments

How I built an open-source AI agent platform that hit 73K GitHub stars — and what I learned about building in public

When I first started Paperclip, I wasn't trying to build a company. I was trying to solve a problem that was driving me crazy.

The problem: I was building AI-powered features and found myself juggling multiple AI agents across different tools — OpenAI, Claude, local models — without any way to orchestrate them. Each agent ran in isolation. There was no state management, no error handling across agents, no way to see what was happening.

So I built Paperclip.

The early days

I started by hacking together a simple orchestrator in Node.js. The idea was straightforward: give developers a way to define, deploy, and manage AI agent teams from a single codebase. Think Kubernetes, but for AI agents.

The first version was rough. Really rough. But I put it on GitHub anyway, wrote a blog post about what I was building, and shared it on Hacker News.

The response was overwhelming. Within 48 hours, the repo had 2,000 stars. Developers were excited about the concept. They were forking it, opening issues, contributing PRs.

The turning point

The real turning point came when I decided to open-source the entire platform. This was a scary decision — would people still pay for something they could run themselves?

Turns out, open-source was the best marketing decision I ever made. Companies started deploying Paperclip internally. Contributors showed up from around the world. The community grew organically because people genuinely wanted to help build something useful.

Building in public taught me

  1. Share your struggles, not just your wins. The posts that resonated most weren't about our successes — they were about the bugs, the design dead-ends, the features I had to kill.

  2. Documentation is a growth engine. Every time I wrote detailed docs for a new feature, traffic spiked. Developers love good docs.

  3. Community > marketing. 90% of our growth came from word-of-mouth. Happy users tell other developers. Build something good, and the community does the marketing for you.

Where we are now

Today, Paperclip is at 73,000+ GitHub stars. We have thousands of developers using the platform to orchestrate AI agents in production. We just launched our hosted cloud platform, and the response has been incredible.

But the most rewarding part isn't the numbers. It's seeing what other developers build with Paperclip — things I never imagined. That's the magic of open-source.

Would love to hear from others building in public. What's worked for you? What hasn't?


Building Paperclip in public at paperclip.ing · GitHub: paperclipai/paperclip

on July 10, 2026
  1. 1

    Congratulations — 73K GitHub stars is a remarkable signal.

    The provider- and model-level cost tracking caught my attention, especially since Paperclip is designed to coordinate agents across different runtimes and providers.

    Do you already support a generic OpenAI-compatible provider through the adapter or plugin layer?

    I’m building AIMOWAY, a reliable OpenAI-compatible AI API access platform backed by officially authorized upstream service access. Paperclip looks like a strong integration fit, and we would be happy to test the provider setup and contribute a concise configuration guide if that would be useful.

  2. 1

    One thing I'd add is that open source didn't just make distribution easier—it made the product easier to trust.

    For infrastructure, adoption is often limited less by features than by confidence. Letting developers inspect the code reduces a different kind of friction than marketing ever could.