Skip to main content

LangExtract

LangExtract is a Python library that uses large language models to extract structured information from unstructured text, following instructions you write. You describe what to pull out and show a few examples; LangExtract runs the model, maps every extraction back to its exact location in the source text, and can render the results as an interactive, highlighted HTML view.

It is built for text that doesn't fit a fixed schema up front (clinical notes, reports, long-form documents), where you want structured output and a way to trace each value back to where it came from.

What it gives you​

  • Source grounding. Every extraction is tied to a character span in the original text. Extractions the model could not locate in the source are flagged (their character interval is empty), so you can filter out hallucinated values.
  • Structured output from examples. You don't define a schema by hand. A few high-quality examples shape the output, and on supported models LangExtract applies schema constraints to keep results consistent.
  • Built for long documents. LangExtract chunks long input, processes the chunks in parallel, and can make multiple passes to improve recall on large inputs.
  • Interactive visualization. Results export to JSONL and render as a self-contained HTML file that highlights every entity in context.
  • Multiple model backends. Google Gemini (the default), OpenAI, and local models via Ollama are built in, with a plugin system for adding others.

Where to go next​

  • Quickstart: extract and visualize your first results in a few minutes.
  • How extraction works: the pipeline from raw text to grounded results.
  • Grounding: how each value is tied back to the source, and how to tell a real extraction from a hallucination.
  • Model backends: how model_id selects Gemini, OpenAI, or a local model.
  • How-to guides: task recipes for long documents, switching models, writing prompts, and supplying credentials.
  • API reference: signatures, parameters, and types.
Project status

LangExtract is open-source under the Apache 2.0 license. As stated in the project's own README, it is not an officially supported Google product. This documentation site is an independent documentation effort and is not official Google documentation.