Label Studio

HumanSignal's labeling server pairs a Django backend with a React front end, giving annotators one browser workspace for text, images, audio, video, HTML and time series. Each project's interface is defined in an XML-like tag configuration, and templates cover common jobs such as classification, named entities, bounding boxes, polygons and transcription. Data comes from uploaded files or synced S3, Google Cloud Storage and Azure buckets, and results export to JSON, COCO, Pascal VOC and more. Linking a model through the separate ML backend SDK enables pre-labeling, online retraining and active learning, while a REST API, Python SDK and webhooks fit it into pipelines. Install with pip on Python 3.10+, with Docker, or with Docker Compose bundling Nginx and PostgreSQL for production; it listens on port 8080 with SQLite by default. The community edition is Apache-2.0 and usable commercially, but role-based access, workspaces, advanced cloud storage auth and some AI assistants are reserved for the paid Starter Cloud and Enterprise tiers. A sensible default for ML teams wanting self-hosted labeling.