Gemini API Cheatsheet: google-genai SDK Reference

# Python SDK — the modern one (Gemini 2.x era)
pip install google-genai     # ≥ 1.0

# Node / TS SDK
npm install @google/genai

# env — picked up automatically by Client()
export GEMINI_API_KEY=AIza...
# or, equivalently, for AI Studio keys:
export GOOGLE_API_KEY=AIza...

# Vertex AI mode uses ADC instead of an API key:
export GOOGLE_GENAI_USE_VERTEXAI=true
export GOOGLE_CLOUD_PROJECT=my-project
export GOOGLE_CLOUD_LOCATION=us-central1

from google import genai
from google.genai import types
from pydantic import BaseModel

client = genai.Client()

class Invoice(BaseModel):
    vendor: str
    total: float
    line_items: list[str]

resp = client.models.generate_content(
    model="gemini-2.5-flash",
    contents="Extract from: ACME, $129.50, 2x widget, 1x cable",
    config=types.GenerateContentConfig(
        response_mime_type="application/json",
        response_schema=Invoice,
    ),
)

invoice: Invoice = resp.parsed  # typed, validated
print(invoice.vendor, invoice.total)

from google import genai
from google.genai import types

client = genai.Client()

# Plain Python function — schema inferred from type hints + docstring
def get_weather(city: str) -> dict:
    """Current weather for a city."""
    return {"temp_c": 31, "city": city}

# Automatic function calling: SDK executes the call and feeds the result back
resp = client.models.generate_content(
    model="gemini-2.5-flash",
    contents="Weather in Mumbai?",
    config=types.GenerateContentConfig(
        tools=[get_weather],
    ),
)
print(resp.text)

# Manual mode — disable auto-execution and inspect the call
resp = client.models.generate_content(
    model="gemini-2.5-flash",
    contents="Weather in Mumbai?",
    config=types.GenerateContentConfig(
        tools=[get_weather],
        automatic_function_calling=types.AutomaticFunctionCallingConfig(disable=True),
    ),
)
for call in resp.function_calls:
    print(call.name, call.args)

from google import genai

client = genai.Client()

# Streaming variant — same call, _stream suffix, chunk iterator
stream = client.models.generate_content_stream(
    model="gemini-2.5-flash",
    contents="Write a haiku about Mumbai monsoons.",
)

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

# Final usage is on the last chunk's usage_metadata
print(f"\n[done · {chunk.usage_metadata.total_token_count} tokens]")

from google import genai
from google.genai import types

client = genai.Client()

# Multi-turn chat with grounding via Google Search
chat = client.chats.create(
    model="gemini-2.5-flash",
    config=types.GenerateContentConfig(
        system_instruction="You are a concise travel assistant.",
        tools=[types.Tool(google_search=types.GoogleSearch())],
        temperature=0.3,
    ),
)

for question in [
    "What's the weather like in Mumbai in July?",
    "And how does that compare to Goa?",
]:
    resp = chat.send_message(question)
    print(f"Q: {question}\nA: {resp.text}\n")

# Citations from grounded search live in candidate metadata
candidate = resp.candidates[0]
if candidate.grounding_metadata:
    for chunk in candidate.grounding_metadata.grounding_chunks:
        print(f"  - {chunk.web.title}: {chunk.web.uri}")