Quickstart: LangChainGo with Google Gemini

Get started by running your first program with LangChainGo and Google Gemini. Gemini models offer large context windows, native multimodality, and configurable "thinking" on the newer generations.

Prerequisites

1. Google AI API Key: Create an API key at Google AI Studio.
2. Go: Download and install Go.

Setup

Before interacting with the Gemini API, you need to set up your API key as an environment variable.

Linux/macOS (bash/zsh)

export GOOGLE_API_KEY="your_google_api_key_here"

Windows (Command Prompt)

set GOOGLE_API_KEY=your_google_api_key_here

Windows (PowerShell)

$env:GOOGLE_API_KEY="your_google_api_key_here"

For permanent setup, add the environment variable to your shell's profile file (~/.bashrc, ~/.zshrc, etc.) or system environment variables on Windows.

Steps

1. Set up your Google AI API Key: Follow the setup instructions above to configure your API key.

2. Run the example: Execute the following command:

go run github.com/vxcontrol/langchaingo/examples/googleai-completion-example@main-vxcontrol

You should see output naming the second person to walk on the Moon (Buzz Aldrin).

Congratulations! You have successfully built and executed your first LangChainGo LLM-backed program using Google Gemini's cloud-based inference.

Selecting a model

The googleai client reads the API key from the GOOGLE_API_KEY environment variable. Pass it explicitly with googleai.WithAPIKey, and choose a model with googleai.WithDefaultModel:

llm, err := googleai.New(ctx,
    googleai.WithAPIKey(os.Getenv("GOOGLE_API_KEY")),
    googleai.WithDefaultModel("gemini-2.5-flash"),
)

Thinking, structured output, and Vertex AI

• Thinking: on thinking-capable Gemini models, control reasoning per call with llms.WithReasoning (Gemini 3.x maps effort to a thinking level; 2.5 uses a token budget). llms.WithReasoningDisabled() turns thinking off where the model allows it.
• Structured output: request schema-constrained JSON with llms.WithStructuredOutput — see the googleai-structured-output-example.
• Vertex AI: to run Gemini through Google Cloud Vertex AI instead of the AI Studio endpoint, use the llms/googleai/vertex package with vertex.New(ctx, googleai.WithCloudProject(...), googleai.WithCloudLocation(...)). Note that llms.WithReasoningDisabled() succeeds on this door only for models whose thinking is off unless asked for, where omitting the instruction is enough to disable it. On a model that thinks by default the door returns an error rather than pretending: the SDK it is pinned to has no field to carry the instruction. Use the AI Studio door when you need to turn thinking off on such a model.

See Configure LLM Providers for the full set of options.

Here is the entire program (from googleai-completion-example):

// Set the GOOGLE_API_KEY env var to your API key taken from ai.google.dev
package main

import (
	"context"
	"fmt"
	"log"
	"os"

	"github.com/vxcontrol/langchaingo/llms"
	"github.com/vxcontrol/langchaingo/llms/googleai"
)

func main() {
	ctx := context.Background()
	apiKey := os.Getenv("GOOGLE_API_KEY")
	llm, err := googleai.New(ctx, googleai.WithAPIKey(apiKey))
	if err != nil {
		log.Fatal(err)
	}

	prompt := "Who was the second person to walk on the moon?"
	answer, err := llms.GenerateFromSinglePrompt(ctx, llm, prompt)
	if err != nil {
		log.Fatal(err)
	}

	fmt.Println(answer)
}