Google (Gemini)

The Google adapter connects opentine to Gemini models via the google-genai SDK. It exposes Gemini tool schemas and reports supports_thinking when thinking appears in the model name. Cloud live validation requires your Google credentials.

Installation

Install opentine with the Google extra to pull in the google-genai SDK.

Terminal
pip install "opentine[google]"

Quick Start

agent.py
1from opentine import Agent
2from opentine.models.google import Google
3
4# Uses GOOGLE_API_KEY env var by default
5model = Google("gemini-2.0-flash")
6
7agent = Agent(model=model, tools=[...])
8run = agent.run_sync("Compare the top 3 JavaScript frameworks")

Constructor

Signature
Google(
    model: str = "gemini-2.0-flash",
    api_key: str | None = None,  # Falls back to GOOGLE_API_KEY env var
)
  • model — The Gemini model identifier. Defaults to "gemini-2.0-flash".
  • api_key — Your Google API key. If not provided, the adapter falls back to the GOOGLE_API_KEY environment variable.

Authentication

You can provide your API key either directly in the constructor or via an environment variable.

auth.py
1from opentine.models.google import Google
2
3# Explicit API key (overrides env var)
4model = Google("gemini-2.0-flash", api_key="AIza...")
5
6# Or set the environment variable
7# export GOOGLE_API_KEY="AIza..."

Properties

properties.py
1from opentine.models.google import Google
2
3model = Google("gemini-2.0-flash")
4
5print(model.name)               # "gemini-2.0-flash"
6print(model.supports_tools)     # True
7print(model.supports_thinking)  # False
  • name — Returns the model identifier passed to the constructor.
  • supports_tools — True in the adapter. Provider and model-specific tool behavior should be validated in your live gate.
  • supports_thinking — True if the model name contains "thinking", False otherwise.

Supported Models

ModelTool SchemaThinking FlagNotes
gemini-2.0-flashExposedNoFast and cost-effective
gemini-2.0-flash-thinkingExposedExposedFlash with extended thinking
gemini-2.0-proYesNoMost capable Gemini model

Any valid Gemini model identifier can be passed to the constructor.

Tool Use

The adapter converts opentine tool schemas to Gemini function declarations. Pass your tools to the Agent and validate the selected Gemini model in your environment.

tool_use.py
1from opentine import Agent
2from opentine.models.google import Google
3from opentine.tools.search import search
4from opentine.tools.web import fetch
5
6model = Google("gemini-2.0-flash")
7
8agent = Agent(
9    model=model,
10    tools=[search, fetch],
11    system="You are a helpful research assistant.",
12)
13
14run = agent.run_sync("Find the latest news about renewable energy")

Thinking Models

The adapter reports supports_thinking=True when the model name contains "thinking". This is a capability flag today; opentine does not automatically add provider-specific thinking controls.

thinking.py
1from opentine.models.google import Google
2
3# supports_thinking is true when "thinking" appears in the model name.
4flash_thinking = Google("gemini-2.0-flash-thinking")
5print(flash_thinking.supports_thinking)  # True
6
7flash = Google("gemini-2.0-flash")
8print(flash.supports_thinking)           # False

SDK Note

The Google adapter uses the newer google-genai SDK, not the older google-generativeai package. The correct SDK is installed automatically when you use the [google] extra.

Terminal
# The Google adapter uses the google-genai SDK (not the older google-generativeai package).
# This is installed automatically with the [google] extra.
pip install "opentine[google]"

# The SDK is imported internally as:
# from google import genai

Next Steps