VECTOR LAB · EMBEDDINGGEMMA 2 · 384D INT8
388 bytes per vector.
Four verification benches, zero cloud dependency.
Interactive test suite for Phase 1 (384d Int8 Gold Radar), Phase 2 (Bilingual & Voice Context Slicer), Phase 3 (Data-Driven Graph Topology Gate), and Phase 4 (Preview Sandbox & Accept Flywheel).
388 B
Vec384Int8 Frame
100%
Gold Recall@1 & @3
0.76 cos
Dedup Threshold
420 B
Snapshot Record
Embedded 384d Int8 benchmark corpus ready · 388 B/vec (7.92x compression) · optional live llama-server (:18991)