About
Local-first, by design.
JMNI Labs is a small, independent lab building AI infrastructure and applications around a single conviction: capable AI belongs on hardware its users control.
What we do
We build deployment kits, cluster tooling, and reference stacks that make large open-weight models practical to run privately on NVIDIA GB10-class hardware — and we build applications, like Recitation, that show what that hardware is for.
Everything we ship is proven on our own fleet first: six GB10 systems (DGX Spark and ASUS GX10), including switchless multi-node rings serving frontier open models.
How we work
- An operator's lab, not slideware. If we haven't run it, we don't sell it.
- Publish or it didn't happen. Benchmarks, build logs, and kits go out in public, alongside every product.
- Data stays home. Our products assume from the first line of code that your information never needs to leave your hardware.
Who we are
JMNI Labs was founded by John Gutierrez — an engineer who has spent years building private inference systems — and is built together with a close circle of developers and collaborators from the open-model and GB10 communities. We're based in the United States and work remote-first.
Where we're headed
Subscription kits for customer-owned clusters. A public release of Recitation. Bigger rings and bigger models. If any of that is your kind of thing, say hello.