Label Sleuth

Label Sleuth is a free, open-source no-code system for text annotation and building text classifiers. It is aimed at domain experts — physicians, lawyers, psychologists, social scientists — who know their text data but have no machine learning background, and who would otherwise need an NLP specialist to turn that data into a working model. The workflow is deliberately plain: you install it locally with Anaconda and pip, start it, and open it in your browser at localhost, where an intuitive labeling UI lets you mark up examples while machine learning models train automatically in the background. Active learning decides which examples are most worth your time next, so you spend fewer labels reaching a usable classifier; the tool then serves real-time predictions on new text as you go. Label Sleuth was initiated by IBM Research together with academic collaborators, and the project's publications come from work in human-computer interaction, NLP and AI. It is extensible by design, so new model architectures or active learning strategies can be plugged in, and it exposes a REST API plus documented model training, active learning and system configuration options. Typical uses documented by the project are identifying contract clauses such as warranty terms in legal documents, flagging bullying content in large sets of text messages, and classifying customer interactions by request type or sentiment. Because it runs locally, your data stays on your own machine — a meaningful advantage for clinical or legal text you cannot upload to a third party. The trade-offs are equally clear: you supply the Python environment, the compute and the maintenance, and help comes from documentation, publications and the community Slack rather than a support contract.