Trust3 Client Library

The trust3_client library allows you to seamlessly integrate Privacera AI Governance into your Langchain projects.

This README provides a step-by-step guide on how to set up and use the library.

Installation

You can install the trust3_client library using pip:

pip3 install trust3_client

Initialization

Start your PAIG-Server and obtain your API key from the AI application configuration.

# Import trust3_client
import trust3_client.client

# Setup Trust3 Shield with API key
trust3_client.client.setup(frameworks=["langchain"], application_config_api_key="<API_KEY>")

Usage

Once you have completed the setup of trust3_client, you can set the user in the context for Privacera Shield to use.

Using context manager

import trust3_client.client

trust3_client.client.setup(frameworks=["langchain"], application_config_api_key="<API_KEY>")

# Set the current user_name in the context
try:
    with trust3_client.client.create_shield_context(username="user"):
        response = llm_chain.run(prompt_text)
except trust3_client.exception.AccessControlException as e:
    # If access is denied, then this exception will be thrown. You can handle it accordingly.
    print(f"AccessControlException: {e}")

Completed example using OpenAI

import os

import trust3_client.client
import trust3_client.exception

from langchain.llms import OpenAI
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain

api_key=os.getenv("OPENAI_API_KEY") # 

# Initialize Trust3 Shield with API key
trust3_client.client.setup(frameworks=["langchain"], application_config_api_key="<API_KEY>")

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 is first President of USA and where did they live?"
llm_chain = LLMChain(prompt=prompt, llm=llm)
try:
   with trust3_client.client.create_shield_context(username=user):
      response = llm_chain.run(prompt_text)
      print(f"LLM Response: {response}")
except trust3_client.exception.AccessControlException as e:
   # If access is denied, then this exception will be thrown. You can handle it accordingly.
   print(f"AccessControlException: {e}")

Additional ways of configuring Trust3 Shield

• Pass the API key as application_config_api_key parameter to the setup function
• Set the environment variable PAIG_APP_API_KEY to your API key
• Place the configuration file in privacera folder relative to where you are running the app (legacy method)
• Set the environment variable PRIVACERA_SHEILD_CONFIG_FILE to the path of the file (legacy method)
• Set the environment variable PRIVACERA_SHEILD_CONFIG_DIR to a folder that contains the file (legacy method)
• Pass the file path as application_config_file parameter to the setup function (legacy method)
• Pass the string contents of the file to the setup function as application_config parameter (legacy method)
• Pass a dict by converting file contents which is in json format and pass to the setup function as application_config parameter (legacy method)

You can create multiple applications

If your application has multiple AI models to be governed, you can create multiple applications as follows:

app1 = trust3_client.client.setup_app(application_config_api_key="<API_KEY_1>")
app2 = trust3_client.client.setup_app(application_config_api_key="<API_KEY_2>")

You can pass the following parameters to the setup_app function:

• Pass the API key as application_config_api_key parameter
• Pass the file path as application_config_file parameter to the setup function (legacy method)
• Pass the string contents of the file to the setup function as application_config parameter (legacy method)
• Pass a dict by converting file contents which is in json format and pass to the setup function as application_config parameter (legacy method)

And then you can pass the application object to the context manager as follows:

with trust3_client.client.create_shield_context(application=app1, username="user"):
    response = llm_chain.run(prompt_text)

Note that you still need to invoke the trust3_client.client.setup() method before calling the trust3_client.client.setup_app(...) method.