Tools we shipped on the Spark.
Each product is a real, runnable tool built on one NVIDIA DGX Spark — with a build-metrics infographic mined from the session transcripts and git log, and a feature tour you can read in a scroll.
Orionfold Cortex
A local memory layer that re-indexes, scores its own recall, and won't ship a rebuild that regresses it.
4.2h to build 1,013 LOC 8 tests
Orionfold Arena — Control Plane
The cockpit stops recording and starts dispatching — re-evaluate models and auto-confirm leaderboard regressions.
2.6h to build 1,762 LOC 35 tests
Orionfold Arena
The cockpit for running, comparing, and scoring local language models on a DGX Spark.
15.4h to build 12,733 LOC 125 tests
The Living Model
A local domain model re-trained on a cadence against a freshening bench — sold on its own lineage delta chart.