Skip to main content
Blindfold is provider-agnostic. The pattern is always the same: tokenize sensitive data, send safe text to your AI provider, then detokenize the response to restore original values.

​
OpenAI

from blindfold import Blindfold
from openai import OpenAI

bf = Blindfold()  # Free local mode, or add api_key="..." for NLP
openai = OpenAI()

# Tokenize PII
safe = bf.tokenize("My name is John Smith, email john@acme.com")

# Send to GPT — PII never reaches OpenAI
response = openai.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": safe.text}]
)

# Restore original data
result = bf.detokenize(response.choices[0].message.content, safe.mapping)
print(result.text)

​
Anthropic Claude

from blindfold import Blindfold
import anthropic

bf = Blindfold()
client = anthropic.Anthropic()

safe = bf.tokenize("My name is John Smith, email john@acme.com")

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{"role": "user", "content": safe.text}]
)

result = bf.detokenize(response.content[0].text, safe.mapping)
print(result.text)

​
Google Gemini

from blindfold import Blindfold
from google import genai

bf = Blindfold()
client = genai.Client()

safe = bf.tokenize("My name is John Smith, email john@acme.com")

response = client.models.generate_content(
    model="gemini-2.5-flash",
    contents=safe.text
)

result = bf.detokenize(response.text, safe.mapping)
print(result.text)

​
Vercel AI SDK

import { Blindfold } from '@blindfold/sdk';
import { generateText } from 'ai';
import { openai } from '@ai-sdk/openai';

const bf = new Blindfold();

const safe = await bf.tokenize("My name is John Smith, email john@acme.com");

const { text } = await generateText({
  model: openai('gpt-4o'),
  prompt: safe.text
});

const result = bf.detokenize(text, safe.mapping);
console.log(result.text);

​
AWS Bedrock

from blindfold import Blindfold
import boto3
import json

bf = Blindfold()
bedrock = boto3.client("bedrock-runtime", region_name="us-east-1")

safe = bf.tokenize("My name is John Smith, email john@acme.com")

response = bedrock.invoke_model(
    modelId="anthropic.claude-sonnet-4-20250514-v1:0",
    body=json.dumps({
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1024,
        "messages": [{"role": "user", "content": safe.text}]
    })
)

body = json.loads(response["body"].read())
result = bf.detokenize(body["content"][0]["text"], safe.mapping)
print(result.text)

​
Azure OpenAI

from blindfold import Blindfold
from openai import AzureOpenAI

bf = Blindfold()
client = AzureOpenAI(
    azure_endpoint="https://your-resource.openai.azure.com",
    api_version="2024-02-15-preview"
)

safe = bf.tokenize("My name is John Smith, email john@acme.com")

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": safe.text}]
)

result = bf.detokenize(response.choices[0].message.content, safe.mapping)
print(result.text)

​
Framework Integrations

LangChain

Official langchain-blindfold package with BlindfoldPIITransformer and RunnableLambda support

Guardrails AI

Official guardrails-blindfold validator for PII protection in Guardrails pipelines

MCP Server

Official Blindfold MCP server for Claude Desktop and any MCP client

​
CrewAI

from blindfold import Blindfold
from crewai import Agent, Task, Crew

bf = Blindfold()

user_input = "Analyze the account for John Smith, SSN 123-45-6789"
safe = bf.tokenize(user_input)

analyst = Agent(
    role="Data Analyst",
    goal="Analyze user accounts",
    backstory="You are a helpful analyst."
)

task = Task(
    description=safe.text,  # PII-free input
    agent=analyst,
    expected_output="Account analysis"
)

crew = Crew(agents=[analyst], tasks=[task])
output = crew.kickoff()

result = bf.detokenize(output.raw, safe.mapping)
print(result.text)

​
Works with any provider

The pattern is always the same regardless of provider: 
from blindfold import Blindfold

bf = Blindfold()  # Free local mode, or add api_key="..." for NLP

# 1. Tokenize PII
safe = bf.tokenize(user_input)
# "Hi, I'm John Smith" → "Hi, I'm <Person_1>"

# 2. Send to any AI provider
response = your_ai_provider.chat(safe.text)

# 3. Restore original data
result = bf.detokenize(response, safe.mapping)

Quickstart

Start protecting PII in 5 minutes — install the SDK, no signup required for local mode