Newslens

NewsLens is a modular Python SDK designed for production-grade news summarization. It provides a unified interface for model inference, supporting local execution, remote API calls, and mock engines for development.

Installation

NewsLens supports optional dependencies based on your specific requirements.

For standard usage:

pip install newslens

To include specific capabilities, use the following extras:

# For local GPU inference
pip install "newslens[local]"

# For MCP server support
pip install "newslens[mcp]"

# For all features
pip install "newslens[all]"

Usage

The package utilizes a strategy pattern, allowing you to swap inference backends without modifying your application logic.

Programmatic API

You can initialize the NewsLens class by injecting a specific engine.

from newslens.core.model import NewsLens
from newslens.core.engines import RemoteEngine

# Configure the engine
engine = RemoteEngine(api_url="https://api.yourdomain.com")
lens = NewsLens(engine=engine)

# Summarize a single article
summary = lens.summarize("Your article text here.")

# Summarize multiple articles concurrently
summaries = await lens.asummarize_batch(["Article 1", "Article 2"])

FastAPI Server

The package includes a production-ready API server with lifespan management.

from fastapi import FastAPI
from newslens.server.api import app
# The server automatically manages the thread pool and model lifecycle

Agent Integration

NewsLens can be deployed as an MCP server to provide summarization capabilities to AI agents.

from newslens.agents.mcp import create_mcp_server
from newslens.core.model import NewsLens

lens = NewsLens(engine=my_engine)
mcp_server = create_mcp_server(lens)
mcp_server.run(transport="stdio")

Configuration

The SDK uses pydantic-settings. You can configure the BASE_MODEL, ADAPTER_PATH, and MAX_WORKERS via environment variables or a configuration object.

Testing

The package includes a comprehensive test suite. Ensure you have installed the dev dependencies:

pip install "newslens[all]"
pytest

License

This project is licensed under the MIT License