Orionfold Cortex
A local memory layer that re-indexes, scores its own recall, and won't ship a rebuild that regresses it.
Your second brain on one DGX Spark, driven end-to-end on its own corpus: re-index from the Arena control plane, score chunk-recall@k against a gold set, gate any rebuild that drops it, and query with a provenance/trust filter — private by construction, on the machine under your desk.
4 surfaces, one recall loop
Drive the recall loop yourself
The preview is recorded on a DGX Spark and runs sidecar-less — no GPU, no backend, nothing phones home. Click rebuild and watch the re-index and RAG-eval jobs drain; the recall gate, coverage report, and provenance-filtered query console are the genuine cuts.
Walk the knowledge pane, score a rebuild against the gold set, query the Second Brain with a trust filter — then run it for real on your own Spark with one command.
How it came together
Orionfold Cortex is the Arena M10 recall layer — 1,013 lines, 8 tests — driven end-to-end on this repo's own corpus in one launch session (~4 hours) driving + measuring the Arena M10 recall layer. The launch story walks every surface, the measured recall@k, and the dogfood bug the drive surfaced.
Orionfold Cortex · ships inside fieldkit
Run it on your own Spark.
Install fieldkit with the arena extra, point it at your corpus, and open the knowledge pane over your own notes, evals, and research — measured, gated, and private.
Install Cortex
$ pip install fieldkit[arena]▌