LangExtract

Google released LangExtract to make LLM information extraction verifiable: every extracted entity is mapped to exact character offsets in the source text, so results can be highlighted and audited instead of trusted blindly. Developers define an extraction task with a prompt and a handful of few-shot examples rather than fine-tuning, and the library enforces reliable structured output through schema constraints on supported models. Long documents are handled with chunking, parallel processing, and optional multi-pass extraction for better recall, and results render to an interactive HTML visualization for review. Gemini is the primary backend, with OpenAI models, Vertex AI batch processing, and fully local inference through Ollama also supported via a custom provider plugin system. Installation is pip install langextract; the project is Apache-2.0 licensed, sits at version 1.6.0, and has accumulated about 37,900 stars since its 2025 release. Teams use it to pull structured records from clinical notes, legal documents, and long reports at auditable quality.