LangChain¶

Privacera’s integration via LangChain is designed to be nearly touch-free. This is facilitated through the use of Privacera's Shield library, which transparently intercepts calls within LangChain, enforcing policies on the original prompt as well as whenever prompts are altered, whether due to Chains or Retrieval Augmented Generation (RAG). The objective is to ensure that policy adherence is seamlessly maintained across all interactions within the application, irrespective of prompt modifications.

Here are the Quick Start options for trying out the integrations with LangChain.

1. Google Colab Notebook: You can experiment with the LangChain integration using a Google Colab notebook. This option only requires a Google account. Google Colab provides a free Jupyter notebook environment where you can run the Privacera SecureChat application.

2. Sample Application: You can download the sample application and run it in your local environment. This option requires Python to be installed locally.

For both options, you'll need to create a Privacera Shield Application in PAIG and download the corresponding configuration file.

Adding AI Application in PAIG¶

As a first step, you need to add your AI Application in PAIG and we will use that application to integrate PAIG. If you already added the Application to the PAIG, then you can skip this step.

To create a new application, go to Paig Navigator > AI Application and click the CREATE APPLICATION button on the right top. This will open a dialog box where you can enter the details of the application.

Downloading Privacera Shield Configuration File¶

The Privacera Shield configuration file is required to initialize the Privacera Shield library. This file can be downloaded from the PAIG portal.

Navigate to Paig Navigator -> AI Applications and select the application you want to download the configuration file for. Click on the DOWNLOAD APP CONFIG button from the right top to download the configuration file.

Using Google Colab Notebook¶

After you have downloaded the Privacera Shield configuration file, you can go to Open In ColabGoogle Colab NoteBook

Using Python Sample Application¶

The following are the prerequisites for trying out with LangChain

• LangChain needs python 3.11 and above

Sample Code¶

If you like to try first and then understand the code later, then here is a sample application you can try it out quickly. The explanation of the code is provided here.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

import os
import paig_client
from paig_client import client as paig_shield_client
from langchain_openai import OpenAI
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain

api_key = os.getenv("OPENAI_API_KEY")  # (1)

# Initialize Privacera Shield
paig_shield_client.setup(frameworks=["langchain"])

llm = OpenAI(openai_api_key=api_key)
template = """Question: {question}

Answer: Let's think step by step."""
prompt = PromptTemplate(template=template, input_variables=["question"])

# Let's assume the user is "testuser"
user = "testuser"
prompt_text = "Who was the first President of USA and where did they live?"
print(f"Prompt: {prompt_text}")
print()

llm_chain = LLMChain(prompt=prompt, llm=llm)
try:
    with paig_shield_client.create_shield_context(username=user):
        response = llm_chain.invoke(prompt_text)
        print(f"LLM Response: {response.get('text')}")
except paig_client.exception.AccessControlException as e:
    # If access is denied, then this exception will be thrown. You can handle it accordingly.
    print(f"AccessControlException: {e}")

1
2
3
4

langchain-community==0.0.17
langchain-openai==0.0.5
langchain==0.1.5
paig_client

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

import os

import paig_client
from paig_client import client as paig_shield_client
from langchain.llms.bedrock import Bedrock
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain

# Inititialize Privacera Shield
paig_shield_client.setup(frameworks=["langchain"])

model_name = "amazon.titan-tg1-large"
region = "us-west-2"
llm = Bedrock(model_id=model_name, region_name=region)

template = """Question: {question}

Answer: Let's think step by step."""
prompt = PromptTemplate(template=template, input_variables=["question"])

# Let's assume the user is "testuser"
user = "testuser"
prompt_text = "Who was the first President of USA and where did they live?"
llm_chain = LLMChain(prompt=prompt, llm=llm)
try:
    with paig_shield_client.create_shield_context(username=user):
        response = llm_chain.run(prompt_text)
        print(f"LLM Response: {response}")
except paig_client.exception.AccessControlException as e:
    # If access is denied, then this exception will be thrown. You can handle it accordingly.
    print(f"AccessControlException: {e}")

It is recommended to use Python's virtual environment to run the sample application. The following steps show how to create a virtual environment and run the sample application. Create a folder where where you want to run the sample. E.g.

1
2

mkdir -p ~/paig-sample
cd ~/paig-sample

Create a virtual environment and activate it

1
2

python3 -m venv venv
source venv/bin/activate

Install the required python packages

1

pip3 install -r requirements.txt

Copy Privacera Shield configuration file to the privacera folder

1

mkdir -p privacera

1

cp ~/Downloads/privacera-shield-SecureChat-config.json privacera/

If you are using OpenAI, then you need to set the OPENAI_API_KEY environment variable or set it in the code.

1

export OPENAI_API_KEY="<your-openai-api-key>"

Run the sample application

1

python3 sample_langchain_integration.py

Check the PAIG Lens Access Audits Now go to PAIG Lens Access Audits to check the prompts and responses for the testuser.

Code Breakup and explanation¶

In your AI Application you need to initialize Privacera Shield library. Once it is initialized, it will automatically embed itself within the LangChain framework and intercept all requests made by user as well as the iterations betweens agents and chains. The following code snippet shows how to initialize the PAIG Library.

Copy the Privacera Shield Application configuration file¶

Create a folder called privacera in your application and copy the Privacera Shield Application configuration file

Install paig_client¶

Privacera's Shield library needs to be first installed. This can be done by running the following command:

1

pip install paig_client

Importing the PAIG Libraries¶

Add the following imports to your application

1
2

import paig_client
from paig_client import client as paig_shield_client

Initializing the PAIG Library¶

Call the setup method to initialize the Privacera Shield library.

1

paig_shield_client.setup(frameworks=["langchain"])

Setting Privacera Shield Context¶

Before calling Langchain, the Privacera Shield context needs to be set. This is primarily to set the user context

1
2
3
4
5
6
7

try:
    with paig_shield_client.create_shield_context(username=user):
        response = llm_chain.invoke(prompt_text)
        print(f"LLM Response: {response.get('text')}")
except paig_client.exception.AccessControlException as e:
    # If access is denied, then this exception will be thrown. You can handle it accordingly.
    print(f"AccessControlException: {e}")

What Next?