GitHub - ploomber/sanitAI: SanitAI is a drop-in proxy for OpenAI's API to detect and remove PII data.

SanitAI

A secure middleware that sanitizes OpenAI API traffic by:

• Automatically detecting and removing Personal Identifiable Information (PII)
• Preserving message context and meaning
• Operating as a drop-in proxy for existing OpenAI integrations

What your user sends:

Hello, my card number is 4111-1111-1111-1111. Call me at (123) 456-7890

What the OpenAI API sees:

Hello, my card number is <VISA-CARD>. Call me at <US-NUMBER>

1. Video tutorials
2. Screenshots
3. Getting started
4. Deployment
5. Features
6. Running on ARM/Apple silicon

Video tutorials

1. Introduction + adding a new rule
2. Improving a rule with the AI agent
3. Using SanitAI with the OpenAI Python package

Screenshots

Dashboard

Playground

Rule creation wizard

Getting started

# get the code
git clone https://github.com/edublancas/sanitAI
cd sanitAI/

# set your OpenAI key (used for the agent that helps you define the PII rules)
export OPENAI_API_KEY=SOME-OPENAI-KEY

# build the docker image
# NOTE: if building using ARM (Apple silicon, read the section at the end)
docker build -t sanitai .

# run
docker run -p 5001:80 --rm --name sanitai \
-e OPENAI_API_KEY=$OPENAI_API_KEY \
sanitai

Then, open: http://localhost:5001/admin/

Login:

• Email: admin@example.com
• Password: admin123

You can test it in the Playground tab. Alternatively, you can run the sample.py script:

export OPENAI_API_KEY=SOME-OPENAI-KEY
pip install openai

python example.py

Then, look at the logs from the server and you'll see that the user sent:

Hello, my card number is 4111-1111-1111-1111. Call me at (123) 456-7890

But the proxy intercepts the request and sends this to OpenAI:

Hello, my card number is <VISA-CARD>. Call me at <US-NUMBER>

This is the source code for example.py:

from openai import OpenAI


# point to the proxy
client = OpenAI(base_url="http://localhost:5001/proxy/v1")

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {
            "role": "user",
            "content": "Hello, my card number is  4111 1111 1111 1111. Call me at (123) 456-7890",
        },
    ],
)

print(response.choices[0].message.content)

Deployment

You can deploy this to any platform that supports containerized applications (e.g. Ploomber Cloud). The app runs on a single container (see the Dockerfile), and spawns the proxy and the UI via supervisord and NGINX.

If you're looking for enterprise support, contact me.

Features

1. Drop-in proxy for OpenAI (no need to change your code)
2. UI to admin and test rules
3. AI agent to add new rules
4. AI agent to fix rules

Running on ARM (Apple silicon)

If you're running on ARM (Apple silicon), uncommend uncomment the line that installs spaCy in the Dockerfile.