Vertex AI SDK (Python)

Warnings

• breaking Install package is google-cloud-aiplatform but import is 'import vertexai'. LLMs consistently confuse this and generate 'from google.cloud import vertexai' or 'pip install vertexai-sdk'. Both are wrong.
  Fix: pip install google-cloud-aiplatform; import vertexai
• breaking vertexai.generative_models, vertexai.language_models, vertexai.vision_models, vertexai.tuning, vertexai.caching deprecated June 24, 2025. Will be removed June 24, 2026.
  Fix: Migrate to google-genai with vertexai=True. See migration guide at cloud.google.com/vertex-ai/generative-ai/docs/deprecations/genai-vertexai-sdk
• breaking Vertex AI does not use API keys. Uses Application Default Credentials (ADC). No api_key parameter. LLMs frequently hallucinate an api_key argument on vertexai.init().
  Fix: Run: gcloud auth application-default login. Or set GOOGLE_APPLICATION_CREDENTIALS to service account JSON path.
• gotcha Two PyPI packages: google-cloud-aiplatform and vertexai. They install the same underlying package. Installing either gives you both namespaces. Pinning both to different versions causes conflicts.
  Fix: Pin only one. Prefer google-cloud-aiplatform for consistency with GCP tooling.
• gotcha vertexai.init() must be called before any Vertex AI SDK calls. Forgetting this causes authentication or project resolution errors.
  Fix: Always call vertexai.init(project='...', location='...') at startup.
• gotcha location parameter is required and must be a valid GCP region. 'us-central1' is most common. Wrong or missing location raises InvalidArgument errors.
  Fix: Always pass location='us-central1' or your preferred region.

Install

• pip install google-cloud-aiplatform Python (installs both google-cloud-aiplatform and vertexai)
• pip install vertexai Python (alternative — same package, different PyPI name)
• pip install 'google-cloud-aiplatform[agent_engines,adk]' Python (with Agent Engine and ADK support)

Imports

• vertexai.init

  import vertexai
  
  vertexai.init(
      project='your-project-id',
      location='us-central1'
  )

  install is google-cloud-aiplatform but import is vertexai — not google.cloud.aiplatform. Both packages coexist but vertexai is the high-level SDK. google.cloud.aiplatform is the low-level GAPIC client.
• vertexai.generative_models (deprecated)

  # Use google-genai for generative AI on Vertex AI
  from google import genai
  
  client = genai.Client(
      vertexai=True,
      project='your-project-id',
      location='us-central1'
  )
  response = client.models.generate_content(
      model='gemini-2.5-flash',
      contents='Explain neural networks'
  )
  print(response.text)

  vertexai.generative_models deprecated June 24, 2025. Removed June 24, 2026. Migrate to google-genai with vertexai=True.
• Application Default Credentials

  import vertexai
  
  # Uses ADC automatically — run: gcloud auth application-default login
  vertexai.init(project='your-project-id', location='us-central1')

  Vertex AI does not use API keys. Uses Google Cloud Application Default Credentials (ADC) or service account JSON. No api_key parameter exists on vertexai.init().

Quickstart

Vertex AI Gemini call using google-genai with vertexai=True (recommended path from 2025 onwards).

# For Vertex AI generative AI — use google-genai
# pip install google-genai
from google import genai

client = genai.Client(
    vertexai=True,
    project='your-gcp-project-id',
    location='us-central1'
)

response = client.models.generate_content(
    model='gemini-2.5-flash',
    contents='What is Vertex AI?'
)
print(response.text)