langextract-provider-openrouter 0.1.2

LangExtract provider plugin for OpenRouter API

LangExtract OpenRouter Provider

A LangExtract provider plugin for OpenRouter.

Installation

# Using uv (recommended)
uv pip install langextract-provider-openrouter

# Or using pip
pip install langextract-provider-openrouter

Quick Start

1. Set your OpenRouter API key:

   export OPENROUTER_API_KEY=your_key_here
   

2. Use it in LangExtract:

   import langextract as lx
   from langextract import data
   
   # LangExtract requires at least one example
   examples = [
       data.ExampleData(
           text="John Doe works at Google",
           extractions=[
               data.Extraction(extraction_class="Person", extraction_text="John Doe"),
               data.Extraction(extraction_class="Company", extraction_text="Google")
           ]
       )
   ]
   
   # Use the 'openrouter/' prefix followed by the OpenRouter model ID
   result = lx.extract(
       text_or_documents="Alice Smith joined Apple",
       prompt_description="Extract people and companies",
       examples=examples,
       model_id="openrouter/google/gemini-2.0-flash-exp:free"
   )
   
   for ext in result.extractions:
       print(f"{ext.extraction_class}: {ext.extraction_text}")
   

Supported Models

This provider supports any model available on OpenRouter by prefixing the model ID with openrouter/. Common examples include:

• openrouter/google/gemini-3-flash-preview
• openrouter/google/gemini-3-pro-preview
• openrouter/google/gemini-2.5-flash-preview-09-2025
• openrouter/Qwen/Qwen2.5-VL-72B-Instruct
• openrouter/openai/gpt-5

Configuration Options

Customize your requests using the provider_options dictionary.

Reasoning Effort

Control the reasoning effort for models that support it.

• Key: effort
• Values: "low", "medium", "high", "minimal" (default)
• Reference: OpenRouter Reasoning Docs

# Create a model with high reasoning effort
config = factory.ModelConfig(
    model_id="openrouter/Qwen/Qwen2.5-VL-72B-Instruct",
    provider_kwargs={
        "provider_options": {
            "effort": "high"
        }
    }
)

Token Usage

The provider logs token usage (prompt, completion, and total) to the console and standard logs (INFO level) after each request.

[OpenRouter] Token Usage - Prompt: 229, Completion: 127, Total: 356

Routing/Provider Options

These options control how OpenRouter acts as a router:

• sort: Strategy for provider selection ("price", "latency", "throughput"). Default: "price".
• ignore: List of provider names to exclude (e.g., ["Parasail"]).
• allow_fallbacks: Boolean to allow falling back to other providers. Default: True.
• data_collection: "allow" or "deny". Default: "allow".

Usage Example (Inference Override)

result = lx.extract(
    # ... inputs ...
    # Example: GPT-5
    model_id="openrouter/openai/gpt-5",
    provider_options={
        "effort": "medium",
        "sort": "throughput",
        "ignore": ["SlowProvider"]
    }
)

Development

cd langextract-openrouter
uv venv --python 3.12
# Install dependencies
uv pip install -e ../langextract
uv pip install -e .
uv pip install pytest requests
# Run tests
uv run pytest

Complete Example

For a complete, runnable example, see example_usage.py in the root directory.

import langextract as lx
from langextract import data

examples = [
    data.ExampleData(
        text="John Doe works at Google",
        extractions=[
            data.Extraction(extraction_class="Person", extraction_text="John Doe"),
            data.Extraction(extraction_class="Company", extraction_text="Google")
        ]
    )
]

result = lx.extract(
    "Alice Smith joined Apple", 
    "Extract people and companies", 
    examples=examples,
    model_id="openrouter/google/gemini-2.0-flash-exp:free"
)
for ext in result.extractions:
    print(f"{ext.extraction_class}: {ext.extraction_text}")