Hello Ultralytics Team,
I am currently using Ultralytics YOLO for object detection and would like to clarify a few licensing requirements under AGPL-3.0.
I have the following questions:
1. Private Dataset Disclosure
Does the AGPL-3.0 license require me to publicly disclose the dataset used for training?
The dataset provider does not permit public disclosure of the dataset. In this situation, can I still use Ultralytics YOLO under the free AGPL-3.0 license? If not, what options are available?
2. Source Code Disclosure Scope in an Embedded System
Our embedded system consists of two separate processes:
The two processes communicate via sockets.
Under the AGPL-3.0 license, if we disclose the complete source code of the B Process, can we keep the source code of the A Process private (not disclosed)?
I would appreciate your clarification on these licensing requirements.
Thank you for your time and support.
Hi @sunghyun, thanks for the clear questions!
1. Private dataset: AGPL-3.0 doesn’t require you to publish a dataset you aren’t allowed to redistribute. Our AGPL-3.0 compliance guide lists training data only when it is needed to run or retrain the project and is redistributable, so your provider’s restriction doesn’t by itself rule out the free AGPL-3.0 option. You would still need to publicly release the entire project under AGPL-3.0, including its source code, training scripts, configurations and fine-tuned model weights, so do check that your provider allows releasing weights trained on its data. If the weights or code must stay private, that’s what the Ultralytics Enterprise License is for.
2. Processes A and B: Releasing only B isn’t enough. AGPL-3.0 compliance covers the complete larger application or system that incorporates YOLO, and splitting it into two processes that talk over sockets doesn’t take A out of that scope. So either both A and B are open-sourced under AGPL-3.0, or an Enterprise License lets you keep the system proprietary, which is the usual route for embedded devices.
AGPL-3.0 is a great fit for projects you can share fully in the open, and the Enterprise License covers everything else.