spacy-llm integrates large language models into the spaCy NLP framework, enabling zero-shot NLP tasks without training data. It provides modular task definitions for prompting and parsing, serializable LLM components for pipelines, and supports both commercial APIs (OpenAI, Anthropic, Cohere, Google PaLM, Azure) and open-source models (Falcon, Dolly, Llama 2, OpenLLaMA). The package turns unstructured LLM responses into robust structured outputs for named-entity recognition, text classification, and other NLP tasks.