Osaurus started because I wanted to build something people would love.
Earlier this year, I built Dinoki—a desktop pixel AI companion for macOS. It began as a personal project, a fun experiment to see what it would feel like if AI came to life as a tiny digital being. Fully autonomous. Wandering your desktop. Saying quirky stuff. People connected with it in a way I didn't expect.
Then we launched Dinoki Premium, and I noticed something that bothered me.
Our users were paying twice. Once for the app, once for the AI inference behind it. That's not a great experience for anyone—not for users, not for developers trying to build something delightful.
I started digging into why this was the case, and found a landscape that actively punishes developers for caring about craft.
The macOS Developer's Dilemma
If you want to build an AI-powered app on macOS today, here's what you're up against.
You need to integrate with multiple AI providers—each with their own SDKs, authentication flows, and billing systems. You need to decide whether to support local models, and if so, which runtime to use. You need to figure out how to let users bring their own API keys without making the UX feel like a developer tool.
And if you actually care about the Mac experience? Most of the tooling available is built cross-platform. Electron wrappers. Web views. Apps that feel like tourists on macOS rather than natives. You can ship faster by compromising on performance, but you'll feel it every time you use your own product.
The alternative is building everything from scratch—your own inference layer, your own provider integrations, your own settings management. Months of work before you even get to the thing that makes your app yours.
Then there's distribution. The Mac App Store has become so fragmented that serious developers increasingly ship outside of it entirely—dealing with notarization, custom update mechanisms, and payment processing just to avoid the approval gauntlet. A review process that's supposed to ensure quality instead slows down critical updates and forces arbitrary compromises.
And Apple itself? Recent macOS releases have been riddled with issues. The quality bar for the platform's own software has been slipping—poor execution on features that should be foundational. It's hard not to notice that the company which once defined premium software experiences has lost some of that focus.
But none of this means macOS users should have to compromise.
People chose this platform because they care about craft. They're still here. They still want software that feels intentional, fast, and native. The gap between what they deserve and what they're getting has never been wider.
Meanwhile, users face their own fragmentation. Juggling subscriptions—$20/month here, $10/month there. Copying API keys between apps. Wondering why their expensive MacBook Pro is running a sluggish Electron app that could have been a website.
This felt wrong to me.
The people who chose macOS deserve AI tools built with the same philosophy that drew them to the platform. And developers who want to build those tools deserve infrastructure that doesn't fight them every step of the way.
So I built Osaurus.
What Osaurus Is
Osaurus is a native macOS LLM server. 10MB. Pure Swift. Built specifically for Apple Silicon.
It's the missing piece that should have existed from the start—a way to run local models, connect to cloud providers, and give developers and tinkerers a foundation that doesn't require stitching together five different tools.
But Osaurus isn't really about the technology. It's about what the technology enables.
I want to see more lovable AI products on macOS. Products built by people who care about craft. Products that don't ask users to make impossible tradeoffs between privacy, performance, and capability.
For that to happen, we need infrastructure that gets out of the way.
What We Believe
Building Osaurus forced me to articulate what I actually care about.
The researchers at Ink & Switch wrote something years ago that crystallized it for me: we've become "borrowers of our own data." Cloud services offer convenience, but at the cost of ownership. When those services shut down—and they always eventually do—the work created with them disappears. No Wayback Machine can restore a sunsetted web application.
AI shouldn't repeat this mistake.
We're watching the next great technological transformation unfold in real time. AI will reshape how we create, work, and think. Something this significant shouldn't be controlled by a handful of companies deciding what's possible, who gets access, and on what terms.
Here's where we landed:
Local-first, not local-only. Your machine should be the source of truth. Run models locally when you want privacy and speed. Reach out to cloud providers when you need more power. The choice should be yours, not ours. Your work should be at your fingertips—no spinners, no "connecting to server," no wondering if your data will be there tomorrow.
The network is optional. Local-first software works offline, by design. Your AI tools should function on an airplane, in a coffee shop with unreliable WiFi, or simply when you don't want to depend on someone else's infrastructure.
You retain ownership and control. The files on your machine are yours. The models you download are yours. We're not building another walled garden—we're building a foundation that respects your ownership of your tools and your work.
Decentralized by design. The internet became transformative because it was open. AI shouldn't be locked behind platform gatekeepers. We're building infrastructure that gives people choices, not dependencies.
Free as in freedom. Osaurus is open source, MIT licensed. Some things should exist as public goods. This is one of them.
Everyone should be able to be tinkerers without being developers.
Where We Are
We're bootstrapped. A small team building something we believe should exist in the world.
There's still a lot of work to do. The vision is bigger than what we've shipped. But the foundation is solid, and we're shipping every week.
If this resonates with you—if you've felt the friction of building or using AI apps on macOS—we'd love for you to be part of this.
Try Osaurus. Break it. Tell us what's missing. Join the community and help us figure out what comes next.
We're not trying to build the next platform monopoly. We're trying to build the infrastructure that lets a thousand lovable products bloom.