langchain-snowflake 0.1.3

An integration package connecting Snowflake and LangChain

langchain-snowflake

This package contains the LangChain integration with Snowflake

Installation

pip install -U langchain-snowflake

You can authenticate by one of the following ways:

1. Configure credentials by setting the following environment variables:

SNOWFLAKE_USERNAME should hold your Snowflake username.

SNOWFLAKE_PASSWORD should hold your Snowflake password.

SNOWFLAKE_DATABASE should hold a Snowflake database to operate in.

SNOWFLAKE_SCHEMA should hold the name of the schema to operate in.

SNOWFLAKE_ROLE should contain the name of the appropriate Snowflake role.

SNOWFLAKE_AUTHENTICATOR should contain the name of Snowflake authentication method, if not using username/password authentication.

Any of these paramaters can also be passed directly into the CortexSearchRetriever constructor. Those not passed will be inferred from the environment. For instance:

from langchain_snowflake import CortexSearchRetriever

search = CortexSearchRetriever(
        role="snowflake_role",
        database="your_db",
        schema="your_schema",
        search_service="your_cortex_search_service_name",
        search_column="search_column_name",
)

Here, role, database, and schema are passed directly, while SNOWFLAKE_USERNAME, SNOWFLAKE_PASSWORD, and SNOWFLAKE_AUTHENTICATOR are inferred from the environment.

If the SNOWFLAKE_AUTHENTICATOR environment variable or authenticator property is set to externalbrowser, the SNOWFLAKE_PASSWORD/password need not be provided. externalbrowser auth will prompt to log in through an browser popup instead.

2. Alternatively, you can pass in a snowflake.snowpark.Session directly into the constructor with the sp_session argument. See the Snowflake docs on how to create such a session.

You may also override the database or schema on the provided session by passing these additional key word arguments into the constructor.

from langchain_snowflake import CortexSearchRetriever
from snowflake.snowpark import Session

# Create a snowflake session
snowflake_session = Session.builder.config(...).create()

search = CortexSearchRetriever(
        sp_session=snowflake_session,
        search_service="your_cortex_search_service_name",
        search_column="search_column_name",
        columns=["col1", "col2", "col3"],

        # If for any reason `snowflake_session` has a different
        # database than the one containing your cortex search service,
        # you can override with key word arguments.

        # database="your_database_override",
        # database="your_database_override"
)

Cortex Search Retriever

CortexSearchRetriever is a Third Party Retriever that allows users to utilize their Cortex Search Service within Langchain.

Given the service name and columns to search, the retriever will return matches found within the dataset with which the Cortex Search Service is associated.

from langchain_snowflake import CortexSearchRetriever

search = CortexSearchRetriever(
        search_service="your_cortex_search_service_name",
        search_column="<search_column_name>",
        columns=["<col1>", "<col2>"],
        filter={"@eq": {"<column>": "<value>"}},
        limit=5
)

query="foo bar"
result = search.invoke(query)
for doc in result:
    print(doc)

The class requires the arguments below: search_service corresponds to the name of your Cortex Search Service.

search_column is the search column of the Cortex Search Service.

columns corresponds to the columns to return in the search. If null or an empty list, only the search_column will be returned.

filter is an optional argument corresponding to any filters that should be utilized when searching.

limit corresponds to the number of search hits to return.