spaCy Documentation

What spaCy is

spaCy describes itself as industrial-strength natural language processing in Python. It has been developed since 2015 by Explosion, is open source under the MIT licence, and its site lists support for more than 75 languages with 84 trained pipelines across 25 of them. These docs cover installation, the usage guides, the API reference, the model catalogue and project templates.

What it does

spaCy processes text through a pipeline: tokenisation, part-of-speech tagging, dependency parsing, lemmatisation, sentence segmentation, named entity recognition, entity linking and text classification. Pipelines can use fast CNN-based components or transformer-based ones, and you can train components on your own labelled data using config files.

Is it worth your time?

Yes, for practical NLP. When you need to process thousands of documents quickly, pull out names, dates and amounts, split sentences reliably or build rule-based matchers alongside a model, spaCy is often the most productive tool. It also complements Hugging Face: many teams fine-tune a transformer and run it inside a spaCy pipeline, or use spaCy for tokenisation and rules around an LLM.

Where to start

1. The free interactive course, Advanced NLP with spaCy, at course.spacy.io, which has 55 exercises.
2. spaCy 101 in the usage docs: the Doc, Token and Span objects and how pipelines work.
3. Linguistic features: tokenisation, entities and the rule-based Matcher.
4. Training pipelines, when you need custom entity types or text categories.

Prerequisites

Python basics. Some understanding of tokens, entities and classification helps, but the course explains them.

Free and paid

The library, models, docs and course are free. Explosion separately sells Prodigy, an annotation tool, and offers custom pipeline services; you do not need either to use spaCy.

Common questions

spaCy or Hugging Face Transformers? They solve different problems. spaCy is a fast, complete text processing pipeline; Transformers gives you pretrained models to fine-tune. Many projects use both.

Does it support Urdu? Check the language support page for your language: tokenisation support and trained pipelines vary by language.

Best for: developers who need reliable, fast NLP pipelines for real volumes of text.