# Semantic Kernel

Call a search constructor, register the resulting `kernel_function` as a plugin function, and let your kernel or agent invoke it.


# Install

``` bash
pip install "serpapi-search-tools[semantic-kernel]"
```

If Semantic Kernel is already installed in your project, use `pip install serpapi-search-tools` instead.

This direct function example only needs a SerpApi key:

``` bash
export SERPAPI_API_KEY="your-serpapi-key"
```


# Run it

``` python
import asyncio

from semantic_kernel import Kernel
from serpapi_search_tools import web_search


async def main():
    kernel = Kernel()
    kernel.add_function(
        plugin_name="search",
        function=web_search(
            allowed_engines=["google", "bing", "duckduckgo"],
        ),
    )
    result = await kernel.invoke(
        plugin_name="search",
        function_name="web_search",
        query="recent Python packaging releases",
        engine="google",
    )
    print(result)


asyncio.run(main())
```

Add the same `search` plugin to your Semantic Kernel agent when you want the model to choose the function.


# Try it

Invoke the function with `recent Python packaging releases` and inspect the first three Google results.


# Customize the search

``` python
privacy_search = web_search(
    allowed_engines=["duckduckgo", "bing"],
    default_engine="duckduckgo",
    include_examples=False,
    name="privacy_search",
)
```

When you change `name`, use that value as `function_name` during direct kernel invocation.