Skip to main content
This guide walks you through installing Genkit, writing your first AI-powered function, and exploring it with the local Developer UI.
1

Get a Google AI API key

Genkit’s Google AI plugin uses the Gemini API. Get a free API key from Google AI Studio.
Set the key as an environment variable before running any Genkit code:
export GOOGLE_GENAI_API_KEY="your-api-key"
The plugin also accepts GEMINI_API_KEY and GOOGLE_API_KEY. You can alternatively pass the key directly as googleAI({ apiKey: '...' }), but using an environment variable is recommended so credentials are never committed to source control.
2

Install dependencies

Create a new Node.js project (or open an existing one), then install Genkit and the Google AI plugin:
npm install genkit @genkit-ai/google-genai
Install the Genkit CLI globally so you can use the Developer UI:
npm install -g genkit-cli
3

Write your first Genkit app

Create src/index.ts with the following content:
import { genkit } from 'genkit';
import { googleAI } from '@genkit-ai/google-genai';

// Initialize Genkit with the Google AI plugin.
// The plugin reads GOOGLE_GENAI_API_KEY (or GEMINI_API_KEY) from the environment.
const ai = genkit({
  plugins: [googleAI()],
});

// Define a flow — a traced, deployable AI function.
const jokeFlow = ai.defineFlow('tellJoke', async (topic: string) => {
  const { text } = await ai.generate({
    model: googleAI.model('gemini-2.5-flash'),
    prompt: `Tell me a short joke about ${topic}.`,
  });
  return text;
});

// Run the flow directly when executed as a script.
const joke = await jokeFlow('software engineers');
console.log(joke);
Run it with tsx (or compile with tsc first):
npx tsx src/index.ts
4

Explore with the Developer UI

The Genkit CLI wraps your app with tracing and launches a local Developer UI where you can run flows, inspect execution traces, and compare model outputs interactively.
genkit start -- npx tsx src/index.ts
The CLI starts your app, then opens the Developer UI at http://localhost:4000. In the UI you can:
  • Run the tellJoke flow against any input without touching your code.
  • Inspect traces to see exactly what was sent to and received from the model.
  • Tweak prompts and compare outputs across different Gemini model variants.
Set GENKIT_ENV=dev in your environment to keep the reflection server running even outside the genkit start command, which is useful during active development.
5

Add structured output (optional)

Genkit can validate model responses against a Zod schema and return typed objects:
import { z } from 'genkit';

const RecipeSchema = z.object({
  title: z.string(),
  ingredients: z.array(z.string()),
  steps: z.array(z.string()),
});

const recipeFlow = ai.defineFlow(
  { name: 'generateRecipe', outputSchema: RecipeSchema },
  async (dish: string) => {
    const { output } = await ai.generate({
      model: googleAI.model('gemini-2.5-flash'),
      prompt: `Create a recipe for ${dish}.`,
      output: { schema: RecipeSchema },
    });
    return output!;
  }
);

const recipe = await recipeFlow('chocolate chip cookies');
console.log(recipe.title, recipe.ingredients);

​
Next steps

Concepts: Flows

Learn how flows add observability, retries, and HTTP exposure to any AI function.

Concepts: Models

Understand model references, config options, multimodal inputs, and streaming.

Guides: Structured output

Return validated, type-safe JSON from any model call using Zod schemas.

Guides: Streaming

Stream tokens to the client as they are generated for a faster perceived response.

Guides: Agents

Build multi-step agentic workflows with tool calling and looping.

Plugins: Google AI

Full reference for the @genkit-ai/google-genai plugin including Vertex AI, Imagen, and embeddings.

Plugins overview

Browse all available plugins: Vertex AI, Ollama, Firebase, and community providers.

Developer tools

Deep dive into the Genkit CLI and Developer UI.