spacy-llm

Why we included this project

Anyone who builds NLP pipelines in spaCy will find this useful when they want LLM output to flow through their existing pipeline rather than sit in a chat window. You get named entity recognition, text classification, lemmatization, and entity linking without curating training data: prompts act as the task definition, and the usual serialization and pipeline guarantees stay intact. The modular model interface means the same task can run on OpenAI, Cohere, Anthropic, Google, or open models from Hugging Face, and even LangChain models work inside the pipeline. That makes it easy to prototype and later switch vendors or move to a self-hosted model without rewriting your NLP code. Long prompts that exceed the context window are handled with map-reduce, which is a nice touch for real workloads.