Local-first AI products
Frontier AI, running on
hardware you own.
JMNI Labs builds the stacks and the applications that keep capable AI local — multi-node inference tooling for NVIDIA GB10 systems, published checkpoints, and apps that never send your data to someone else's cloud.
What we build
Infrastructure
Local AI infrastructure
Deployment kits, cluster tooling, and benchmarked reference stacks that turn NVIDIA GB10 systems (DGX Spark, ASUS GX10) into private, multi-node model servers — no network switch required.
Explore infrastructure → ApplicationRecitation
A private AI professor. Upload a textbook; Recitation builds the course — curriculum, interactive lectures, graded problem sets — and runs entirely on your machine.
Meet Recitation →6GB10 systems in the lab
2–4node switchless rings
NVFP4multi-node serving
93Bcheckpoint published