Generative AI Google Brain

gemini-getting-started-intro-gemini-express

3 min read

Getting started with Gemini using Vertex AI in Express Mode

Source notebook

Repo path: gemini/getting-started/intro_gemini_express.ipynb · Open on GitHub · intro

Introduces Gemini on Vertex AI Express Mode using the Google Gen AI SDK.

Summary

This notebook teaches how to call Gemini through Vertex AI in Express Mode with an API key. It demonstrates text generation, streaming, chat, async calls, model configuration, system instructions, safety settings, multimodal inputs, controlled JSON output, token counting, Google Search grounding, function calling, and code execution.

Key code patterns

Create Express Mode client

from google import genai
 
client = genai.Client(
    enterprise=True,
    api_key=API_KEY,
)

Authenticates to Vertex AI Express Mode using an API key.

Generate text

response = client.models.generate_content(
    model=MODEL_ID,
    contents="What's the largest planet in our solar system?",
)
 
response.text

Shows the basic request-response pattern for Gemini text generation.

Stream content

for chunk in client.models.generate_content_stream(
    model=MODEL_ID,
    contents="Tell me a story...",
):
    output_text += chunk.text

Streams chunks as they are generated to reduce perceived latency.

Multi-turn chat

chat = client.chats.create(model=MODEL_ID)
response = chat.send_message("Write a function...")
response = chat.send_message("Write a unit test...")

Preserves conversation context across turns.

Configure generation

config=GenerateContentConfig(
    temperature=0.4,
    top_p=0.95,
    top_k=20,
    max_output_tokens=100,
    seed=5,
)

Controls sampling, length, reproducibility, and stopping behavior.

Send multimodal input

response = client.models.generate_content(
    model=MODEL_ID,
    contents=[
        Part.from_bytes(data=image, mime_type="image/png"),
        "Write a short blog post...",
    ],
)

Combines media parts and text prompts in one Gemini request.

Controlled JSON output

response = client.models.generate_content(
    model=MODEL_ID,
    contents="List cookie recipes.",
    config=GenerateContentConfig(
        response_mime_type="application/json",
        response_schema=Recipe,
    ),
)

Constrains model output to a declared schema and exposes parsed results.

Tool use

google_search_tool = Tool(google_search=GoogleSearch())
response = client.models.generate_content(
    model=MODEL_ID,
    contents="When is the next total solar eclipse...",
    config=GenerateContentConfig(tools=[google_search_tool]),
)

Adds Google Search as a grounding tool that Gemini can use at runtime.

Models & APIs used

  • Models: gemini-3.5-flash
  • APIs / services: Vertex AI, Gemini API
  • SDKs / libraries: google-genai, pydantic

When to use this

Use this pattern to quickly prototype Gemini features on Vertex AI with an API key before building a fuller application.

Gotchas & caveats

  • An API key is required for Vertex AI Express Mode, either inline or from GOOGLE_API_KEY.
  • compute_tokens is only supported in Vertex AI.
  • Public URLs are required when sending web pages by URI.
  • Lower media_resolution can reduce processing time and cost but may affect output quality.
  • Safety response text can be None when content is blocked, with finish_reason set to SAFETY.

Best practices

  • Use streaming responses when reducing perceived latency matters.
  • Use system instructions to steer model behavior for a task.
  • Inspect finish_reason and safety_ratings when using safety settings.
  • Use response_schema and response_mime_type for structured JSON output.
  • Use count_tokens before sending a request to estimate input token usage.
  • Set temperature to 0 for deterministic tool and function calling examples.

Graph View

Backlinks