Google Gemini

Google's google-genai SDK uses httpx for HTTP requests. Rdakt AI integrates via the http_options parameter.

Setup

import httpx
from google import genai
from rdakt_ai import RdaktMiddleware

client = genai.Client(
    http_options={
        "transport": RdaktMiddleware(inner=httpx.AsyncHTTPTransport()),
    },
)

Gemini's SDK passes the transport via http_options dictionary, not through an httpx.AsyncClient instance like OpenAI and Anthropic.

Supported Models

All Gemini generative models: Gemini 2.0 Flash, Gemini 2.5 Flash, Gemini 2.5 Pro, and future models.

Message Format

Gemini uses a different format from OpenAI and Anthropic:

{
  "contents": [
    {
      "parts": [{"text": "Tell me about John Smith"}],
      "role": "user"
    }
  ]
}

Rdakt AI detects Gemini requests by the URL (generativelanguage.googleapis.com) and scans contents[].parts[].text for PII.

Streaming Format

Gemini SSE events contain:

{
  "candidates": [
    {
      "content": {
        "parts": [{"text": "Hello"}],
        "role": "model"
      }
    }
  ]
}

Rdakt AI extracts content from candidates[0].content.parts[0].text.

Streaming Example

stream = await client.aio.models.generate_content_stream(
    model="gemini-2.0-flash",
    contents="Write about Jane Doe (jane@acme.com)",
)

async for chunk in stream:
    print(chunk.text, end="", flush=True)

Non-Streaming Example

response = await client.aio.models.generate_content(
    model="gemini-2.0-flash",
    contents="Summarize the case for John Smith (SSN 123-45-6789)",
)
print(response.text)

Key Differences from OpenAI/Anthropic

Feature	OpenAI/Anthropic	Gemini
Transport injection	`httpx.AsyncClient(transport=...)`	`http_options={"transport": ...}`
Message field	`messages[].content`	`contents[].parts[].text`
Response content	`choices[0].delta.content`	`candidates[0].content.parts[0].text`
Gzip compression	Anthropic only	No