Vertex AI Integration

Route your Google Vertex AI (Gemini) traffic through Rivaro for runtime enforcement. GCP project/location URL structure, Bearer token authentication, and the Gemini content format.

SDK Configuration

Python

import google.auth
from google.auth.transport.requests import Request
import requests

credentials, project = google.auth.default()
credentials.refresh(Request())

RIVARO_BASE = "https://your-org.rivaro.ai"
PROJECT_ID = "my-gcp-project"
LOCATION = "us-central1"
MODEL = "gemini-pro"

url = (
    f"{RIVARO_BASE}/v1/projects/{PROJECT_ID}/locations/{LOCATION}"
    f"/publishers/google/models/{MODEL}:generateContent"
)

response = requests.post(
    url,
    headers={
        "Authorization": f"Bearer {credentials.token}",
        "Content-Type": "application/json",
        "X-Detection-Key": "detect_live_your_key_here"
    },
    json={
        "contents": [{
            "role": "user",
            "parts": [{"text": "Hello, world!"}]
        }],
        "generationConfig": {
            "temperature": 0.7,
            "maxOutputTokens": 1024
        }
    }
)

print(response.json()["candidates"][0]["content"]["parts"][0]["text"])

curl

curl "https://your-org.rivaro.ai/v1/projects/my-project/locations/us-central1/publishers/google/models/gemini-pro:generateContent" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "X-Detection-Key: detect_live_your_key_here" \
  -d '{
    "contents": [{
      "role": "user",
      "parts": [{"text": "Hello"}]
    }]
  }'

Supported Endpoints

Endpoint	Method	Description
/v1/projects/{project}/locations/{location}/publishers/google/models/{model}:generateContent	POST	Generate content
/v1/projects/{project}/locations/{location}/publishers/google/models/{model}:streamGenerateContent	POST	Generate with streaming

Request and response formats match the Vertex AI Gemini API exactly.

URL Structure

Vertex AI endpoints include your GCP project and location in the path:

/v1/projects/{projectId}/locations/{location}/publishers/google/models/{model}:generateContent

Component	Example	Description
projectId	my-gcp-project	Your GCP project ID
location	us-central1	GCP region
model	gemini-pro	Model name

Authentication

Vertex AI uses GCP Bearer tokens, not API keys:

Authorization: Bearer <gcp-access-token>

Generate a token using gcloud auth print-access-token or programmatically via the Google Auth library. Rivaro passes the Authorization header through to Vertex AI unchanged.

Required Headers

Header	Required	Description
X-Detection-Key	Yes	Your Rivaro detection key
Authorization	Yes	Bearer <gcp-access-token>
Content-Type	Yes	application/json

Request Format

Vertex AI uses the Gemini content format with contents and parts:

{
  "contents": [
    {
      "role": "user",
      "parts": [{"text": "Explain quantum computing"}]
    }
  ],
  "generationConfig": {
    "temperature": 0.7,
    "maxOutputTokens": 2048,
    "topP": 0.95,
    "topK": 40
  }
}

Response format

{
  "candidates": [{
    "content": {
      "role": "model",
      "parts": [{"text": "Quantum computing uses..."}]
    },
    "finishReason": "STOP"
  }],
  "usageMetadata": {
    "promptTokenCount": 10,
    "candidatesTokenCount": 85,
    "totalTokenCount": 95
  }
}

Streaming

Use streamGenerateContent for streaming responses:

curl "https://your-org.rivaro.ai/v1/projects/my-project/locations/us-central1/publishers/google/models/gemini-pro:streamGenerateContent" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "X-Detection-Key: detect_live_your_key_here" \
  -d '{
    "contents": [{"role": "user", "parts": [{"text": "Hello"}]}]
  }'

Content chunks are forwarded in real time. Enforcement runs on the accumulated response after the stream completes (finishReason: "STOP" or finishReason: "MAX_TOKENS").

Function Calling

Vertex AI function calling works through the proxy. Function calls appear in response parts as functionCall objects:

{
  "candidates": [{
    "content": {
      "parts": [{
        "functionCall": {
          "name": "get_weather",
          "args": {"city": "London"}
        }
      }]
    }
  }]
}

Rivaro inspects these function call parts against detection rules.

AppContext Configuration

When creating an AppContext for Vertex AI, the configuration map supports:

Key	Description
projectId	GCP project ID
location	GCP region (default: us-central1)
model	Model name (default: gemini-pro)

Next steps

• Error Handling — Handle Rivaro-specific errors
• Understanding Detections — What Rivaro scans for
• API Reference — Full endpoint reference