How to do per-user retrieval

This guide demonstrates how to configure runtime properties of a retrieval chain. An example application is to limit the documents available to a retriever based on the user.

When building a retrieval app, you often have to build it with multiple users in mind. This means that you may be storing data not just for one user, but for many different users, and they should not be able to see eachother's data. This means that you need to be able to configure your retrieval chain to only retrieve certain information. This generally involves two steps.

Step 1: Make sure the retriever you are using supports multiple users

At the moment, there is no unified flag or filter for this in LangChain. Rather, each vectorstore and retriever may have their own, and may be called different things (namespaces, multi-tenancy, etc). For vectorstores, this is generally exposed as a keyword argument that is passed in during similarity_search. By reading the documentation or source code, figure out whether the retriever you are using supports multiple users, and, if so, how to use it.

Note: adding documentation and/or support for multiple users for retrievers that do not support it (or document it) is a GREAT way to contribute to LangChain

Step 2: Add that parameter as a configurable field for the chain

This will let you easily call the chain and configure any relevant flags at runtime. See this documentation for more information on configuration.

Now, at runtime you can call this chain with configurable field.

Code Example

Let's see a concrete example of what this looks like in code. We will use Pinecone for this example.

To configure Pinecone, set the following environment variable:

PINECONE_API_KEY: Your Pinecone API key

from langchain_openai import OpenAIEmbeddings
from langchain_pinecone import PineconeVectorStore

embeddings = OpenAIEmbeddings()
vectorstore = PineconeVectorStore(index_name="test-example", embedding=embeddings)

vectorstore.add_texts(["i worked at kensho"], namespace="harrison")
vectorstore.add_texts(["i worked at facebook"], namespace="ankush")

API Reference:OpenAIEmbeddings | PineconeVectorStore

['ce15571e-4e2f-44c9-98df-7e83f6f63095']

The pinecone kwarg for namespace can be used to separate documents

# This will only get documents for Ankush
vectorstore.as_retriever(search_kwargs={"namespace": "ankush"}).get_relevant_documents(
    "where did i work?"
)

[Document(page_content='i worked at facebook')]

# This will only get documents for Harrison
vectorstore.as_retriever(
    search_kwargs={"namespace": "harrison"}
).get_relevant_documents("where did i work?")

[Document(page_content='i worked at kensho')]

We can now create the chain that we will use to do question-answering over.

Let's first select a LLM.

pip install -qU langchain-openai

import getpass
import os

os.environ["OPENAI_API_KEY"] = getpass.getpass()

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o-mini")

pip install -qU langchain-anthropic

import getpass
import os

os.environ["ANTHROPIC_API_KEY"] = getpass.getpass()

from langchain_anthropic import ChatAnthropic

llm = ChatAnthropic(model="claude-3-5-sonnet-20240620")

pip install -qU langchain-openai

import getpass
import os

os.environ["AZURE_OPENAI_API_KEY"] = getpass.getpass()

from langchain_openai import AzureChatOpenAI

llm = AzureChatOpenAI(
    azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
    azure_deployment=os.environ["AZURE_OPENAI_DEPLOYMENT_NAME"],
    openai_api_version=os.environ["AZURE_OPENAI_API_VERSION"],
)

pip install -qU langchain-google-vertexai

# Ensure your VertexAI credentials are configured

from langchain_google_vertexai import ChatVertexAI

llm = ChatVertexAI(model="gemini-1.5-flash")

pip install -qU langchain-aws

# Ensure your AWS credentials are configured

from langchain_aws import ChatBedrock

llm = ChatBedrock(model="anthropic.claude-3-5-sonnet-20240620-v1:0",
    beta_use_converse_api=True)

pip install -qU langchain-cohere

import getpass
import os

os.environ["COHERE_API_KEY"] = getpass.getpass()

from langchain_cohere import ChatCohere

llm = ChatCohere(model="command-r-plus")

pip install -qU langchain-nvidia-ai-endpoints

import getpass
import os

os.environ["NVIDIA_API_KEY"] = getpass.getpass()

from langchain_nvidia_ai_endpoints import ChatNVIDIA

llm = ChatNVIDIA(model="meta/llama3-70b-instruct")

pip install -qU langchain-fireworks

import getpass
import os

os.environ["FIREWORKS_API_KEY"] = getpass.getpass()

from langchain_fireworks import ChatFireworks

llm = ChatFireworks(model="accounts/fireworks/models/llama-v3p1-70b-instruct")

pip install -qU langchain-groq

import getpass
import os

os.environ["GROQ_API_KEY"] = getpass.getpass()

from langchain_groq import ChatGroq

llm = ChatGroq(model="llama3-8b-8192")

pip install -qU langchain-mistralai

import getpass
import os

os.environ["MISTRAL_API_KEY"] = getpass.getpass()

from langchain_mistralai import ChatMistralAI

llm = ChatMistralAI(model="mistral-large-latest")

pip install -qU langchain-openai

import getpass
import os

os.environ["TOGETHER_API_KEY"] = getpass.getpass()

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    base_url="https://api.together.xyz/v1",
    api_key=os.environ["TOGETHER_API_KEY"],
    model="mistralai/Mixtral-8x7B-Instruct-v0.1",
)

This is basic question-answering chain set up.

from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import (
    ConfigurableField,
    RunnablePassthrough,
)

template = """Answer the question based only on the following context:
{context}
Question: {question}
"""
prompt = ChatPromptTemplate.from_template(template)

retriever = vectorstore.as_retriever()

API Reference:StrOutputParser | ChatPromptTemplate | ConfigurableField | RunnablePassthrough

Here we mark the retriever as having a configurable field. All vectorstore retrievers have search_kwargs as a field. This is just a dictionary, with vectorstore specific fields.

This will let us pass in a value for search_kwargs when invoking the chain.

configurable_retriever = retriever.configurable_fields(
    search_kwargs=ConfigurableField(
        id="search_kwargs",
        name="Search Kwargs",
        description="The search kwargs to use",
    )
)

We can now create the chain using our configurable retriever

chain = (
    {"context": configurable_retriever, "question": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

We can now invoke the chain with configurable options. search_kwargs is the id of the configurable field. The value is the search kwargs to use for Pinecone

chain.invoke(
    "where did the user work?",
    config={"configurable": {"search_kwargs": {"namespace": "harrison"}}},
)

'The user worked at Kensho.'

chain.invoke(
    "where did the user work?",
    config={"configurable": {"search_kwargs": {"namespace": "ankush"}}},
)

'The user worked at Facebook.'

For more vectorstore implementations for multi-user, please refer to specific pages, such as Milvus.

Code Example​

Was this page helpful?

Code Example