Licenses

OpenProcessor itself

OpenProcessor's own code is licensed AGPL-3.0-or-later. It was previously MIT-badged and was re-licensed because it vendors an AGPL-3.0 Ultralytics fork (src/ultralytics_patches/, from levipereira/ultralytics, for end2end TensorRT export), and AGPL's copyleft terms propagate to the combined work. Third-party components keep their own licenses. The authoritative per-component record is ATTRIBUTION.md in the repository.

Model weights

Code licenses and weight licenses differ, and several weights are not redistributable. This project does not ship those weights; you download them yourself and are bound by their terms directly. Verify each model card before any commercial deployment.

Model (Triton name)	Weights source	Weight license	Redistributable?
YOLO11s end2end (yolov11_small_trt_end2end)	Ultralytics release assets	AGPL-3.0	Yes, under AGPL
YOLO26 (yolo26_small_trt, trainer bases yolo26s.pt, yolo26n.pt)	Ultralytics	AGPL-3.0	Yes, under AGPL
SCRFD-10G (scrfd_10g_bnkps), ArcFace w600k_r50 (arcface_w600k_r50)	InsightFace model zoo (buffalo_l)	Non-commercial research only	No, you download them
MobileCLIP2-S2 image/text	apple/MobileCLIP2-S2 on HuggingFace	Apple ML Research (apple-amlr), research-only	No, you download them
PE-Core-L14-336 (pe_image_encoder, pe_text_encoder)	facebook/PE-Core-L14-336 on HuggingFace (not gated)	Apache-2.0	Yes
PP-OCRv5 det/rec (paddleocr_det_trt, paddleocr_rec_trt, ocr_pipeline)	PaddleOCR, via third-party ONNX repackaging	Apache-2.0	Yes
SAM 3 (segmenter)	facebook/sam3 on HuggingFace (gated, needs HF_TOKEN)	SAM License	Yes with the SAM License attached, but gated
VLM (bring your own)	Whatever OpenAI-compatible model you configure	Depends on the model	Check the model card you deploy

Notes:

• InsightFace: the MIT license covers InsightFace's code. The buffalo_l pretrained weights used for SCRFD and ArcFace are published for non-commercial research use only. For commercial face features, swap in commercially licensed detector and embedding weights.
• MobileCLIP: the ml-mobileclip code is under the Apple Sample Code License; the MobileCLIP2-S2 weights are under apple-amlr, a research-only term. make export-mobileclip fetches them on your machine.
• SAM 3: the code is Apache-2.0; the weights are under Meta's SAM License (commercial use and redistribution allowed with the license attached) and gated on HuggingFace. See docker/segmenter/NOTICE.

Software components

Component	License
NVIDIA Triton Inference Server	BSD 3-Clause
NVIDIA TensorRT	NVIDIA software license
Ultralytics YOLO, levipereira/ultralytics fork	AGPL-3.0
InsightFace code (SCRFD, ArcFace)	MIT
Apple MobileCLIP code	Apple Sample Code License
OpenCLIP	MIT
PaddleOCR	Apache-2.0
Meta Perception Encoder (PE-Core) code	Apache-2.0
Meta SAM 3 code	Apache-2.0
OpenSearch	Apache-2.0

Sample datasets

Fetched on demand by make sample-coco, make sample-coco-readme, make sample-coco-cars, make sample-coco-import and make sample-plates; nothing is bundled in the repository. The car and import subsets use only CC BY images, and the wheel boxes in the import fixture's region layouts are synthetic geometry derived from COCO car boxes, not annotations.

Dataset	Annotations	Images
COCO 2017	CC BY 4.0 (COCO Consortium)	Per-image Flickr licenses, filtered to Attribution, Attribution-ShareAlike, "No known copyright restrictions" and "United States Government Work"; per-image credit in the fetched ATTRIBUTION.csv
Open Images V7	CC BY 4.0 (Google LLC)	Listed CC BY 2.0 and verified per image; per-image credit in the fetched ATTRIBUTION.csv
Ultralytics test images (bus.jpg, zidane.jpg)		AGPL-3.0 (make download-test-images)

Citations: Lin, T.-Y. et al. (2014), "Microsoft COCO: Common Objects in Context," ECCV; Kuznetsova, A. et al. (2020), "The Open Images Dataset V4," IJCV.

These docs

The screenshots in this documentation were captured from Cropwright, the labeling frontend in this repository, running against public sample data only. See Screenshots and image credits.