LangExtract is a Python library for using large language models to extract structured information from unstructured documents. Users describe the desired entities or relationships with instructions and a small number of examples, and the library turns model output into structured extractions tied back to the exact source text.

A major focus is grounding and reviewability. LangExtract records character locations for extracted content, can flag results that cannot be located in the source, and can generate an interactive HTML visualization for inspecting extracted entities in context. It also supports long documents through chunking, parallel processing, and multiple passes, and can work with cloud models such as Gemini or local models through Ollama.

The library is useful for document-processing tasks in domains such as healthcare, legal analysis, research, and business operations where structured data must be derived from free text but traceability matters. Its few-shot design allows new extraction schemas to be defined without model fine-tuning, while source alignment makes it easier to verify results and build human-review workflows around LLM extraction.