LangChain

The langchain-expunct package is drop-in PII redaction middleware for LangChain. Add one line to your chain and automatically strip sensitive data before it reaches any LLM.

Package: langchain-expunct on PyPI

Installation


pip install langchain-expunct

Get your API key at expunct.ai — free tier includes 1M tokens/month.

Quick Start

Redact text in a chain


from langchain_openai import ChatOpenAI
from langchain_expunct import ExpunctPIIRedactor
 
redactor = ExpunctPIIRedactor(api_key="pk_live_...")
llm = ChatOpenAI()
 
chain = redactor | llm
result = chain.invoke("Summarize this: John Smith (SSN 219-09-9999) applied for a loan.")
# The LLM sees: "Summarize this: PERSON_1 (SSN US_SSN_1) applied for a loan."

Redact chat messages


from langchain_core.messages import HumanMessage, SystemMessage
from langchain_openai import ChatOpenAI
from langchain_expunct import ExpunctChatMessageRedactor
 
redactor = ExpunctChatMessageRedactor(api_key="pk_live_...")
chat = ChatOpenAI()
 
chain = redactor | chat
result = chain.invoke([
    SystemMessage(content="You are a helpful assistant."),
    HumanMessage(content="My name is Alice Johnson and my email is alice@example.com"),
])

Async usage


import asyncio
from langchain_openai import ChatOpenAI
from langchain_expunct import ExpunctPIIRedactor
 
redactor = ExpunctPIIRedactor(api_key="pk_live_...")
llm = ChatOpenAI()
chain = redactor | llm
 
async def main():
    result = await chain.ainvoke("Call Bob at 415-555-0100")
    print(result)
 
asyncio.run(main())

Callback handler — auto-redact LLM inputs


from langchain_openai import ChatOpenAI
from langchain_expunct import ExpunctCallbackHandler
 
handler = ExpunctCallbackHandler(api_key="pk_live_...", redact_input=True, log_output=True)
llm = ChatOpenAI()
 
result = llm.invoke(
    "Tell me about John Smith, SSN 219-09-9999",
    config={"callbacks": [handler]},
)

For async workflows, use AsyncExpunctCallbackHandler:


from langchain_expunct import AsyncExpunctCallbackHandler
 
handler = AsyncExpunctCallbackHandler(api_key="pk_live_...")
result = await llm.ainvoke("...", config={"callbacks": [handler]})

Components

`ExpunctPIIRedactor`

A Runnable[str, str] that redacts PII from plain text. Composable with |.

Parameter	Type	Default	Description
`api_key`	`str`	required	Your Expunct API key
`base_url`	`str`	`https://api.expunct.ai`	API base URL
`policy_id`	`str \| None`	`None`	Policy to apply
`language`	`str`	`"en"`	Language of input text

`ExpunctChatMessageRedactor`

A Runnable[list[BaseMessage], list[BaseMessage]] that redacts PII from messages while preserving roles, names, and metadata.

Parameter	Type	Default	Description
`api_key`	`str`	required	Your Expunct API key
`base_url`	`str`	`https://api.expunct.ai`	API base URL
`language`	`str`	`"en"`	Language of input text

`ExpunctCallbackHandler` / `AsyncExpunctCallbackHandler`

Hooks into on_llm_start and on_llm_end to redact inputs and log detected PII in outputs.

Parameter	Type	Default	Description
`api_key`	`str`	required	Your Expunct API key
`base_url`	`str`	`https://api.expunct.ai`	API base URL
`language`	`str`	`"en"`	Language of input text
`redact_input`	`bool`	`True`	Redact PII from prompts on LLM start
`log_output`	`bool`	`True`	Log detected PII in LLM responses

RAG Pipeline Example


from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_community.vectorstores import FAISS
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from langchain_expunct import ExpunctPIIRedactor
 
vectorstore = FAISS.from_texts(["..."], embedding=OpenAIEmbeddings())
retriever = vectorstore.as_retriever()
 
redactor = ExpunctPIIRedactor(api_key="pk_live_...")
prompt = ChatPromptTemplate.from_template(
    "Answer based on context: {context}\n\nQuestion: {question}"
)
llm = ChatOpenAI()
 
rag_chain = (
    {"context": retriever, "question": redactor}
    | prompt
    | llm
    | StrOutputParser()
)
 
result = rag_chain.invoke("What did John Smith order? His email is john@acme.com")