| Kodama Vision: locally adapted decision model. | |
| Pretrained ModernBERT: Answer.AI, LightOn, and collaborators. Apache-2.0 per the included upstream model card. | |
| Frozen SigLIP: Google. Apache-2.0 per the included upstream model card. | |
| Changes: typed-decision and escalation heads, local fine-tuning, calibration, image adapter and similarity prior. | |
| Original Kodama additions and runtime: no separate public license selected for this release. | |