Pushing Local AI Forward
For the past three months I’ve been building and shipping local AI recipes, mostly for NVIDIA DGX Spark systems.
What started as a personal project has grown far beyond what I expected. My GitHub repositories are now being used by people around the world to deploy local AI models more easily, with a strong focus on performance, stability, and reliability.
One example is my DeepSeek v4 Flash recipe for a two node DGX Spark cluster. It has become one of most widely used recipes in the world for a cluster of two DGX Sparks, and, has even made its way into local LLM study groups in Tokyo.
"Mia, your recipe is now being used all over the world! Thank you for the wonderful work"
This repo alone has over 1,300 stars on GitHub and 21 contributors.
And this isn’t an isolated case. Many of my recipes have hundreds of stars, and those numbers continue to grow.
The journey has been intense, but incredibly rewarding. I’ve learned a lot, met some amazing people, and become even more convinced that local AI has a bright future.
What's next
I have big plans. My mission is simple: make local AI more accessible to everyone while delivering the best performance and stability the hardware can support.
To do that, two things need to happen.
1. Build software people can depend on
The projects I release need to be actively maintained and continuously improved.
I don’t want to publish one-off repositories, move on to the next thing, and leave users behind. I want to build software that people can depend on, projects that keep getting faster, more stable, and more capable over time.
2. Bring local AI to more hardware
Although most of my recipes are for NVIDIA DGX Sparks, this is not enough to bring true change.
Great local AI experiences should be available across a much broader range of devices, including consumer NVIDIA GPUs, Apple Silicon, AMD hardware, Strix Halo systems, and more.
The challenge is that I quickly realized there simply aren’t enough hours in the day to do all of this alone.
So I started looking for like-minded people who share the same vision and want to help build it.
Not long ago, @plotarmordev joined me to work on DeepSeek v4 Flash. Together, we made it significantly more stable and reliable. Today, I’m comfortable calling it rock-solid and production-ready.
And since then, the team has continued to grow.
We are now 10 people, including me.
It's still early, and will take time to refine how we work together, figure out where each person can have the most impact, and reach our full potential as a team.
But I’m incredibly excited about what we can build together. This is no longer just about maintaining a few projects.
It’s about building the software, tooling, and infrastructure that can make high-quality local AI practical across as much hardware as possible.
But We Need Help.
I’m raising funding to support the team and give us the tools and hardware we need to keep building, maintaining, testing, and expanding these projects.
Funding will help us buy and access hardware, maintain existing releases, test across more platforms, run benchmarks, improve documentation, and give contributors more time to work on the project.
We would also greatly appreciate hardware donations, especially:
• Apple Silicon systems
• AMD Strix Halo systems
If you believe in making powerful local AI accessible to more people, please consider sponsoring our work on GitHub.
Every contribution helps us spend more time building, testing, optimizing, maintaining, and supporting open local AI.
GitHub Sponsors:
github.com/sponsors/MiaAI-Lab
You can also support us with crypto:
A big thank you to everyone who has used our work, contributed code, reported issues, shared feedback, donated hardware, or simply supported and encouraged us along the way.
And a special thank you to everyone who has joined me in the mission to make local AI better - @plotarmordev@u1tra_instinct@jvr0x@mrexodia@PalazzoVincenzo@jauneclaude@CMichaelGibbs@Italianclownz