spaCy

Built by Explosion AI, it provides pre-trained pipelines for named entity recognition (NER), part-of-speech tagging, dependency parsing, text classification, and sentence segmentation across over 70 languages.

Unlike research-focused NLP libraries, spaCy makes deliberate trade-offs in favor of speed and API clarity, making it the standard choice for engineering teams shipping NLP features into real products.

spaCy's architecture centers on the Doc object a rich data structure that stores a tokenized text alongside all annotations produced by the pipeline. This design enables efficient batch processing and easy integration with downstream components.

The library supports transformer-based models (via spacy-transformers) for state-of-the-art accuracy, as well as smaller statistical models for low-latency applications.

Custom components can be added to the pipeline declaratively, allowing teams to extend spaCy with domain-specific logic medical entity recognition, legal clause extraction, financial sentiment classification without rewriting the processing infrastructure.

spaCy is open-source under the MIT license and maintained by Explosion AI alongside a large community of contributors. It integrates natively with Hugging Face models, Prodigy (a data annotation tool), and Weaviate for vector search workflows.

Organizations including major financial institutions, healthcare providers, and technology companies use spaCy in production NLP systems processing billions of documents.