LangExtract GuideMaster Google's AI Data Extraction Library

Why Trust This Guide? ​

Official documentation is great, but real-world projects are messy. As developers using LangExtract in production, we built this guide to bridge the gap between "Hello World" and deployment.

We cover local LLM setups, cost optimization, and handling complex documents — things you won't find in the standard README.

Quick Start ​

Get up and running in 30 seconds.

1. Install ​

Install via pip. Requires Python 3.9+.

pip install langextract

2. Configure API Key ​

By default, LangExtract uses Google Gemini. Get your key from Google AI Studio.

export LANGEXTRACT_API_KEY="your-api-key-here"

3. Your First Extraction ​

Extract characters from a simple text using few-shot examples.

import langextract as lx

# Define your extraction prompt
prompt = "Extract characters and their emotions from the text."

# Provide a high-quality example to guide the model
examples = [
    lx.data.ExampleData(
        text="ROMEO. But soft! What light through yonder window breaks?",
        extractions=[
            lx.data.Extraction(
                extraction_class="character",
                extraction_text="ROMEO",
                attributes={"emotional_state": "wonder"}
            ),
        ]
    )
]

# Input text to process
text = "Lady Juliet gazed longingly at the stars, her heart aching for Romeo."

# Run extraction (uses Gemini Flash by default)
result = lx.extract(
    text_or_documents=text,
    prompt_description=prompt,
    examples=examples,
    model_id="gemini-2.5-flash",
)

print(result.extractions)
print(result.extractions)

4. Visualize the Results (Interactive Visualization) 📊 ​

LangExtract's killer feature. Generate an interactive HTML report for easy verification.

# 1. Save as JSONL
lx.io.save_annotated_documents([result], output_name="extraction_results.jsonl", output_dir=".")

# 2. Generate interactive HTML
html_content = lx.visualize("extraction_results.jsonl")
with open("visualization.html", "w") as f:
    f.write(html_content.data if hasattr(html_content, 'data') else html_content)

View Visualization Example

LLM Configuration Guide ​

LangExtract supports multiple backends. Here is how to configure them.

Local LLMs with Ollama 🏠 ​

Great for privacy and zero cost.

1. Install Ollama: ollama.com
2. Pull a Model: ollama pull gemma2:2b
3. Run Ollama Server: ollama serve
4. Configure Code:

import langextract as lx

result = lx.extract(
    text_or_documents=text,
    prompt_description=prompt,
    examples=examples,
    model_id="gemma2:2b",  # Automatically selects Ollama provider
    model_url="http://localhost:11434",
    fence_output=False,
    use_schema_constraints=False
)

OpenAI GPT-4 🧠 ​

Best for complex reasoning tasks. Requires optional dependency: pip install langextract[openai]

export OPENAI_API_KEY="sk-..."

import os
import langextract as lx

result = lx.extract(
    text_or_documents=text,
    prompt_description=prompt,
    examples=examples,
    model_id="gpt-4o",  # Automatically selects OpenAI provider
    api_key=os.environ.get('OPENAI_API_KEY'),
    fence_output=True,
    use_schema_constraints=False
)

OpenAI-Compatible APIs 🔌 ​

LangExtract works with any OpenAI-compatible API, including DeepSeek, Qwen, Doubao, and more.

import langextract as lx

result = lx.extract(
    text_or_documents=text,
    prompt_description=prompt,
    examples=examples,
    model_id="deepseek-chat",
    api_key="your-api-key",
    language_model_params={
        "base_url": "https://api.deepseek.com/v1"  # Replace with provider's URL
    },
    fence_output=True,
    use_schema_constraints=False
)
    use_schema_constraints=False
)

Google Vertex AI Batch (Enterprise) 🏢 ​

For large-scale tasks, enable Batch mode to save costs.

result = lx.extract(
    ...,
    language_model_params={"vertexai": True, "batch": {"enabled": True}}
)

Advanced Installation: Docker 🐳 ​

Run without polluting your local environment:

docker run --rm -e LANGEXTRACT_API_KEY="your-key" langextract python your_script.py

Scaling to Longer Documents 📚 ​

How to handle books or PDFs larger than the context window? LangExtract features built-in Chunking and Parallel Processing.

No need to split text manually. Just pass the URL or long text:

# Example: Process the entire "Romeo and Juliet"
result = lx.extract(
    text_or_documents="https://www.gutenberg.org/files/1513/1513-0.txt",
    prompt_description=prompt,
    examples=examples,
    # Core parameters
    extraction_passes=3,    # Multiple passes to improve recall
    max_workers=20,         # Parallel workers for speed
    max_char_buffer=1000    # Context buffer size
)

Real-World Data Extraction Examples ​

Explore our collection of Google AI data extraction examples for common use cases.

🏥 Medical Report Extraction ​

Extract medication names, dosages, and frequencies from clinical notes. View Full Example

📚 Long Text Extraction ​

Handling PDFs or books that exceed token limits. View Full Example

🇯🇵 Multilingual Extraction ​

Working with non-English text (Japanese, Chinese). View Full Example

FAQ ​

Q: What is the difference between LangExtract and Docling? A: Docling specializes in parsing documents (like PDFs) into Markdown, handling layout analysis. LangExtract focuses on extracting structured data (like JSON) from text. They work great together: use Docling to parse, and LangExtract to structure the data.

Q: Can I use DeepSeek, Groq, or other OpenAI-compatible models? A: Absolutely. LangExtract supports any model with an OpenAI-compatible API. Just set the base_url to your provider's endpoint. It works seamlessly with DeepSeek V3/R1, Groq, local vLLM, etc.

Q: How do I handle documents longer than the context window? A: LangExtract has built-in chunking mechanisms. Check out our Long Text Extraction Example to see how it automatically splits long texts, processes them (in parallel or sequence), and merges the results.

Q: Can I run this locally for privacy? A: Yes. Integrate with Ollama to run models like Llama 3 or Mistral locally. This is free and ensures data never leaves your machine, ideal for medical or legal data.

Q: Is LangExtract free? A: The library is 100% open-source and free. You only pay for the LLM API usage (e.g., Google Gemini, OpenAI). If run locally with Ollama, it operates completely cost-free.