Unlock Structured Data from Any Text with LangExtract – A Free Python LLM Tool

Recently discovered a super‑useful Python tool on GitHub – LangExtract, an open‑source text‑extraction library (Apache 2.0) that turns messy documents into structured data with just a few lines of code.

What is LangExtract?

LangExtract is a Python library that extracts structured information (names, dates, relationships, etc.) from unstructured text such as medical records, contracts, news, or novels. It leverages Google’s Gemini model (or a local Ollama model) and works by providing a natural‑language prompt and a few examples.

Super simple : No regex or model training required.

Precise provenance : Returns the original text positions, useful for audit scenarios.

Handles long documents : Processes millions of characters with smart chunking and multithreading.

Cool visualisation : Generates an HTML page with highlighted results.

Free and open source : Community‑driven and customizable.

Code examples

Four scenarios are demonstrated: extracting medication and dosage from medical records, clauses and dates from contracts, characters and relationships from a novel, and company names with industries from news articles.

Example 1 – Medications from medical records

Scenario : Extract drug names and dosages from a batch of doctor notes.

import langextract as lx
import textwrap

# Define the task
prompt = textwrap.dedent("""
Extract medication and dosage from the text.
Example:
Text: Patient takes Lisinopril 10mg daily, Metformin 500mg twice a day.
Output:
- Medication: Lisinopril, Dosage: 10mg
- Medication: Metformin, Dosage: 500mg
""")

text = "患者每天吃 Lisinopril 10mg 治高血压，Metformin 500mg 每天两次治糖尿病。"

result = lx.extract(text, prompt_description=prompt, model_id="gemini-2.5-flash")
lx.io.save_annotated_documents([result], "meds.jsonl")
html = lx.visualize("meds.jsonl")
with open("meds.html", "w") as f:
    f.write(html)

print(result.extractions)

Result (JSONL) :

[
  {"entity": "Lisinopril", "type": "药物", "dosage": "10mg", "start_char": 8, "end_char": 19},
  {"entity": "Metformin", "type": "药物", "dosage": "500mg", "start_char": 28, "end_char": 37}
]

Opening meds.html shows the extracted entities highlighted in the original text.

Example 2 – Clauses and dates from contracts

Scenario : Pull clause titles and effective dates from legal contracts.

import langextract as lx
import textwrap

prompt = textwrap.dedent("""
Extract clause names and effective dates.
Example:
Text: Rental clause effective 2023‑01‑01, renewal clause effective 2024‑01‑01.
Output:
- Clause: Rental clause, Effective date: 2023‑01‑01
- Clause: Renewal clause, Effective date: 2024‑01‑01
""")

text = "保密条款2025年3月1日生效，违约条款2025年4月1日生效。"

result = lx.extract(text, prompt_description=prompt, model_id="gemini-2.5-flash")
lx.io.save_annotated_documents([result], "contract.jsonl")
html = lx.visualize("contract.jsonl")
with open("contract.html", "w") as f:
    f.write(html)

print(result.extractions)

Result (JSONL) :

[
  {"entity": "保密条款", "type": "条款", "effective_date": "2025年3月1日", "start_char": 0, "end_char": 4},
  {"entity": "违约条款", "type": "条款", "effective_date": "2025年4月1日", "start_char": 15, "end_char": 19}
]

Opening contract.html highlights the clauses and dates.

Example 3 – Characters and relationships from a novel

Scenario : Analyse character relationships in "Romeo and Juliet".

import langextract as lx
import textwrap

prompt = textwrap.dedent("""
Extract characters and their relationships.
Example:
Text: Romeo loves Juliet, Juliet is a Capulet.
Output:
- Character: Romeo, Relation: loves, Target: Juliet
- Character: Juliet, Relation: belongs to, Target: Capulet family
""")

text = """
罗密欧深爱朱丽叶，但蒙太古家族和凯普莱特家族是世仇。
朱丽叶的表哥提伯尔特跟罗密欧打了一架。
"""

result = lx.extract(
    text,
    prompt_description=prompt,
    model_id="gemini-2.5-flash",
    max_workers=4,
    num_passes=2
)

lx.io.save_annotated_documents([result], "romeo.jsonl")
html = lx.visualize("romeo.jsonl")
with open("romeo.html", "w") as f:
    f.write(html)

print(result.extractions)

Result (JSONL) :

[
  {"entity": "罗密欧", "type": "人物", "relation": "爱上", "target": "朱丽叶", "start_char": 0, "end_char": 3},
  {"entity": "朱丽叶", "type": "人物", "relation": "属于", "target": "凯普莱特家族", "start_char": 7, "end_char": 10},
  {"entity": "提伯尔特", "type": "人物", "relation": "冲突", "target": "罗密欧", "start_char": 20, "end_char": 25}
]

Opening romeo.html visualises the characters and their relationships.

Example 4 – Companies and industries from news

Scenario : Extract company names and their industries from financial news.

import langextract as lx
import textwrap

prompt = textwrap.dedent("""
Extract company names and industries.
Example:
Text: Apple releases new iPhone, Tesla expands battery production.
Output:
- Company: Apple, Industry: Technology
- Company: Tesla, Industry: Automotive/Energy
""")

text = "华为发布新芯片，字节跳动进军AI教育。"

result = lx.extract(text, prompt_description=prompt, model_id="gemini-2.5-flash")
lx.io.save_annotated_documents([result], "news.jsonl")
html = lx.visualize("news.jsonl")
with open("news.html", "w") as f:
    f.write(html)

print(result.extractions)

Result (JSONL) :

[
  {"entity": "华为", "type": "公司", "industry": "科技", "start_char": 0, "end_char": 2},
  {"entity": "字节跳动", "type": "公司", "industry": "科技/AI", "start_char": 10, "end_char": 14}
]

Opening news.html highlights the companies and their industries.

How to get started in 5 minutes

Clone the repository:

git clone https://github.com/google/langextract.git && cd langextract

Install dependencies:

pip install -e ".[dev]"  # for development
pip install -e ".[test]"  # for testing

Set the API key (required for Gemini):

echo 'LANGEXTRACT_API_KEY=YOUR_KEY' >> .env && echo '.env' >> .gitignore

Run any of the examples above, adjust the text and prompt, and view the generated HTML visualisation.

Tips and caveats

Medical use : LangExtract is a demonstration tool, not a diagnostic system.

Cost : Using Gemini may incur API fees; local Ollama models are free.

Large documents : Use Tier 2 quota to avoid rate‑limiting.

Conclusion

LangExtract acts as a smart assistant that converts chaotic text into tidy tables and visualisations, making it valuable for healthcare, legal, literary, and financial workflows.