bioscope-langextract 0.0.1

A lightweight library for extracting structured data from language models

LangExtract (Lightweight Fork)

A lightweight fork of Google's langextract optimized for serverless deployment (AWS Lambda, Cloud Functions, etc.).

What's Different in This Fork?

This fork removes heavy dependencies and unused features:

Removed	Reason
numpy, pandas	Only used in visualization/benchmarks and CSV reading
google-genai	Users provide their own providers
google-cloud-storage	Batch processing removed
absl-py	Replaced with stdlib logging
tqdm	Progress bars removed
requests, aiohttp	URL fetching removed

Key changes:

• No built-in providers (Gemini, OpenAI, Ollama) - you register your own
• No visualization module
• No URL fetching - pass text directly
• No progress bars
• Stdlib logging with optional logger injection

Installation

pip install bioscope-langextract

Or with uv:

uv add bioscope-langextract

Quick Start

1. Create a Custom Provider

Since this fork has no built-in providers, you must register your own:

import langextract as lx
from langextract.core.base_model import BaseLanguageModel
from langextract.core.types import ScoredOutput
from langextract.providers import router

@router.register(r"^my-model", priority=100)
class MyProvider(BaseLanguageModel):
    """Custom provider that wraps your LLM API."""

    def infer(self, batch_prompts, **kwargs):
        # Implement your LLM call here
        for prompt in batch_prompts:
            response = call_your_llm(prompt)  # Your implementation
            yield [ScoredOutput(score=1.0, output=response)]

2. Define Your Extraction Task

import textwrap

prompt = textwrap.dedent("""\
    Extract characters, emotions, and relationships in order of appearance.
    Use exact text for extractions. Do not paraphrase or overlap entities.
    Provide meaningful attributes for each entity to add context.""")

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"}
            ),
        ]
    )
]

3. Run the Extraction

input_text = "Lady Juliet gazed longingly at the stars, her heart aching for Romeo"

result = lx.extract(
    text_or_documents=input_text,
    prompt_description=prompt,
    examples=examples,
    model_id="my-model",  # Matches your registered pattern
)

4. Optional: Inject a Custom Logger

For serverless environments, you can inject your own logger:

import logging

logger = logging.getLogger("my-app")

result = lx.extract(
    text_or_documents=input_text,
    prompt_description=prompt,
    examples=examples,
    model_id="my-model",
    logger=logger,
)

Provider Registration

Register providers using regex patterns:

from langextract.providers import router

# Decorator style
@router.register(r"^claude-", r"^anthropic/", priority=100)
class ClaudeProvider(BaseLanguageModel):
    ...

# Or register after definition
router.register(r"^gpt-")(OpenAIProvider)

# Lazy registration (defers import)
router.register_lazy(
    r"^bedrock/",
    target="my_package.providers:BedrockProvider",
    priority=50
)

Higher priority wins when multiple patterns match.

API Reference

lx.extract()

Main extraction function:

lx.extract(
    text_or_documents,      # str or list of documents
    prompt_description,     # Task description
    examples,               # List of ExampleData
    model_id,               # Model ID (matched against registered patterns)
    *,
    extraction_passes=1,    # Number of extraction passes
    max_workers=1,          # Parallel workers for chunked processing
    max_char_buffer=4000,   # Chunk size for long documents
    logger=None,            # Optional custom logger
    **kwargs
)

Data Classes

# Example for few-shot prompting
lx.data.ExampleData(
    text="...",
    extractions=[...]
)

# Single extraction
lx.data.Extraction(
    extraction_class="category",
    extraction_text="exact text from source",
    attributes={"key": "value"}
)

License

This is a modified fork of langextract by Google LLC, licensed under the Apache License 2.0. This fork removes heavy dependencies (numpy, pandas, google-genai, etc.) for serverless deployment environments.

See LICENSE for the full license text.

Disclaimer

This is not an officially supported Google product. Use is subject to the Apache 2.0 License.