Prodigy Recipes

Prodigy is a modern annotation tool for creating high-quality training and evaluation data for machine learning models. It goes beyond simple labeling with workflows for data inspection, cleaning, error analysis, and rule-based system development. Built as a downloadable Python library with a web app, it supports named entity recognition, text classification, span categorization, computer vision, audio/video annotation, and prompt engineering. Users define custom data feeds and interfaces, making annotation over 10x more efficient by automating repetitive tasks. Prodigy runs entirely on your machines—no data ever leaves—and the models you produce are yours with no lock-in. It integrates tightly with the spaCy ecosystem and offers lifetime licensing with predictable pricing. The tool is designed for data scientists, NLP engineers, and ML practitioners who need to actively engage with data to build robust models. Key features include pre-built annotation workflows, custom recipe scripts (Python), active learning to prioritize informative examples, LLM integration for assisted labeling (e.g., ner.llm.correct), and a plugin system (SSO, Modal, DSPy). Recent updates (v1.18.6–v1.18.7) fixed bugs in audio.manual and custom logo display. Prodigy's privacy-first, self-hosted architecture sets it apart from cloud-only annotation tools.