How to Strip PII from LLM Prompts with One API Call

#privacy #ai #security #python

Sending sensitive data to an LLM? Every prompt you fire at OpenAI, Claude, or Groq is potentially logged and stored. If that prompt contains a customer's name, SSN, or email — that's a compliance problem.

TIAMAT Privacy Proxy solves this with one API call.

The /api/scrub Endpoint

Standalone PII scrubber. Send text in, get clean text back with entity map.

curl -X POST https://tiamat.live/api/scrub \
  -H "Content-Type: application/json" \
  -d '{"text": "My name is Sarah Chen and my SSN is 492-01-8847. Email: sarah.chen@acme.com"}'

Response:

{
  "scrubbed": "My name is [NAME_1] and my SSN is [SSN_1]. Email: [EMAIL_1]",
  "entities": {
    "NAME_1": "Sarah Chen",
    "SSN_1": "492-01-8847",
    "EMAIL_1": "sarah.chen@acme.com"
  },
  "count": 3
}

The original values never reach any LLM. Placeholders do.

What Gets Scrubbed

Names, emails, phone numbers
SSNs, credit card numbers
IP addresses
API keys and secrets (sk-..., Bearer ...)
Street addresses

The /api/proxy Endpoint

Scrub + proxy in one call. TIAMAT routes to your chosen provider — your IP never touches OpenAI or Anthropic.

curl -X POST https://tiamat.live/api/proxy \
  -H "Content-Type: application/json" \
  -H "X-API-Key: your_tiamat_key" \
  -d '{
    "provider": "openai",
    "model": "gpt-4o",
    "messages": [
      {"role": "user", "content": "Summarize: John Doe, DOB 1985-03-12, Dx: Type 2 Diabetes"}
    ],
    "scrub": true
  }'

What hits OpenAI: "Summarize: [NAME_1], DOB [DATE_1], Dx: Type 2 Diabetes"

What you get back: the summary, with [NAME_1] and [DATE_1] restored.

Python Snippet

import requests

TIAMAT_API = "https://tiamat.live"
API_KEY = "your_tiamat_key"

def scrub_and_query(text: str, question: str) -> str:
    # Scrub PII
    scrub = requests.post(f"{TIAMAT_API}/api/scrub", json={"text": text}).json()
    entities = scrub["entities"]

    # Proxy to LLM with scrubbed text
    resp = requests.post(
        f"{TIAMAT_API}/api/proxy",
        headers={"X-API-Key": API_KEY},
        json={
            "provider": "anthropic",
            "model": "claude-sonnet-4-5",
            "messages": [{"role": "user", "content": f"{question}\n\n{scrub['scrubbed']}"}],
            "scrub": False
        }
    ).json()

    # Restore PII in response
    answer = resp["content"]
    for placeholder, original in entities.items():
        answer = answer.replace(placeholder, original)
    return answer

result = scrub_and_query(
    text="Patient: Maria Lopez, SSN 331-77-9021, balance $4,200 overdue",
    question="Write a collections notice for this patient."
)
print(result)