diffgram

Why we included this project

Diffgram is a self-hosted system for teams that want their training data, schemas, and model predictions in one place rather than scattered across separate tools. It pairs a labeling interface that handles images, video, audio, text, 3D, and conversational data with a workflow layer for organizing datasets and moving them toward training, so you avoid stitching together an annotation tool, a storage bucket, and a pipeline on your own. Because it runs on your own infrastructure, it suits organizations that cannot send sensitive material to a cloud vendor. One caveat: the project uses a custom Diffgram license v2 plus a Contributor License, not a standard open-source grant, so read the terms before relying on it. It earns its keep in annotation-heavy workflows where the built-in human supervision tools do most of the heavy lifting.