Quickstart: OpenAI

Prerequisites

rdakt-ai and openai installed
OPENAI_API_KEY environment variable set

Async Usage

import httpx
from openai import AsyncOpenAI
from rdakt_ai import RdaktMiddleware

client = AsyncOpenAI(
    http_client=httpx.AsyncClient(
        transport=RdaktMiddleware(inner=httpx.AsyncHTTPTransport()),
    ),
)

response = await client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "user", "content": "Summarize the case for John Smith (SSN 123-45-6789)"}
    ],
)
print(response.choices[0].message.content)

Sync Usage

import httpx
from openai import OpenAI
from rdakt_ai import RdaktSyncMiddleware

client = OpenAI(
    http_client=httpx.Client(
        transport=RdaktSyncMiddleware(inner=httpx.HTTPTransport()),
    ),
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "user", "content": "Summarize the case for John Smith (SSN 123-45-6789)"}
    ],
)
print(response.choices[0].message.content)

Streaming

Streaming works automatically. Tokens are deanonymized as they arrive:

stream = await client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "user", "content": "Write a report about Jane Doe, email jane@acme.com"}
    ],
    stream=True,
)

async for chunk in stream:
    content = chunk.choices[0].delta.content
    if content:
        print(content, end="", flush=True)

The LLM sees anonymized values like <PERSON_1> and <EMAIL_1>. Rdakt AI replaces them back to "Jane Doe" and "jane@acme.com" in the streamed response before your code sees it.

What Just Happened?

Detection — Rdakt AI scanned the outbound message and found "John Smith" (PERSON), "123-45-6789" (SSN)
Anonymization — The message sent to OpenAI contained <PERSON_1> and <SSN_1> instead of real values
Deanonymization — The streamed response had tokens replaced back to original values in real-time