AI SDK Dev Setup

You've grasped some of the core LLM concepts. Now for the exciting part: setting up your local development environment and getting hands-on with the Vercel AI SDK. This is where you really get your hands dirty and prepare to build the cool AI features we've been talking about, like automatic summarization and building intelligent chatbots.

You will get the starter project running locally, securely configure the Vercel AI Gateway, and confirm everything works so you're ready to start coding AI in the next lesson.

Let's go!

Prerequisites Check

Before starting, ensure you have Git, Node.js (v20+), and pnpm installed. If you missed any, check the prerequisites section at the beginning of the course for installation links.

Step 1: Getting the Code

First, grab the starter project code. This repository contains all the files and setup needed for the course. Open your terminal (or command prompt) and run this command to clone it:

bash

This copies the project files to your computer.

Step 2: Navigating Into the Project

Now, move into the project directory you just cloned. Use the cd (change directory) command:

bash

Your terminal prompt should now show you're inside the build-an-ai-app-starter-sep-25 folder.

Step 3: Installing Dependencies

Next, install the necessary libraries (AI SDK, Next.js, etc.). This project uses pnpm for fast and efficient package management. Run:

bash

Using npm or Yarn?

While pnpm is recommended for this project (due to the lockfile), you can try using npm install or yarn install. Be aware you might encounter slight differences if dependency versions vary. To install pnpm globally, run: npm install -g pnpm.

This might take a minute or two.

Step 4: Setting Up the Vercel AI Gateway

AI models are accessed through API endpoints and typically this means you need to setup individual accounts and secure API keys for each provider. The Vercel AI Gateway simplifies this and allows you to access models from multiple providers through a single endpoint.

Why Use AI Gateway?

Free $5 credit: Get started without needing an OpenAI account or dealing with their credit requirements

One endpoint, multiple providers: Switch between OpenAI, Anthropic, Google, and others without code changes

Automatic failover: If one provider fails, requests automatically retry with another

Cost tracking: Monitor spending across all providers in one place

Rate limit handling: Built-in retry logic and request queuing

Bring your own keys: Use your existing provider API keys if needed

You have two options for authentication with the AI Gateway:

Option A: OIDC Token Setup (Recommended for Projects Already Deployed to Vercel)

This method uses secure OIDC federation and is ideal for projects deployed to Vercel.

1. Install the Vercel CLI

Install the Vercel CLI globally:

bash

This will allow you to use the vc command in your terminal and manage your Vercel projects.

2. Deploy your Project

Deploying your project will both create and deploy your project to Vercel:

bash

Note that if you have an existing project already deployed to Vercel, you can use the vc link command to vc link your project.

3. Enable OIDC and Get your Token

Once your project is deployed, enable OIDC in your project settings:

Visit your project in the Vercel dashboard
Navigate to Settings → Security
Enable "Secure Backend Access with OIDC Federation"

Then get your token by running:

bash

Check .env.local to see your OIDC token:

bash

OIDC tokens are only valid for 12 hours. You have two options:

Manual refresh: Run vc env pull again to get a new token

Automatic refresh: Use vc dev to run your application locally - it automatically refreshes the OIDC token when it runs

Option B: API Key Setup (Simpler for Local Development)

If you prefer using a persistent API key instead of OIDC tokens, you can create one directly in the Vercel dashboard:

1. Create an API Key

Go to your Vercel dashboard
Navigate to the AI Gateway tab
Click "API keys" in the sidebar
Click "Create key" and follow the dialog
Copy your new API key

2. Configure your Environment

Create or update your .env.local file:

text

Which method should I use?

OIDC (Option A): Best for production deployments and team projects. More secure but requires token refresh.

API Key (Option B): Simpler for local development and learning. Persistent but requires careful key management.

Step 5: Making Sure It Works!

Let's verify your setup. The project has a simple script (env-check.ts) to confirm your authentication is configured correctly. Run it:

bash

You should see this in your terminal:

If using OIDC (Option A):

text

If using API Key (Option B):

text

Token/Key Management

OIDC tokens: Need refreshing every 12 hours. Run vc env pull again or use vc dev for automatic refresh

API keys: Persistent and don't expire, but keep them secure and rotate periodically for best security practices

Setup Checklist

Project repository cloned and dependencies installed?
.env.local file created and API key configured?
env-check.ts ran successfully and showed your key?
Ready to start building AI features?

Common Setup Issues & Troubleshooting

If you get an error about the git command not found, install Git. See the official Git installation guide.
If you get an error about the pnpm command not found, install pnpm globally. See the official pnpm installation guide.
If you get an error about the Node.js version, install Node.js. See the official Node.js installation guide.
env-check.ts shows undefined or errors: Check your .env.local file name (leading dot, no extra extensions). For OIDC, the key name must be VERCEL_OIDC_TOKEN. For API Key, it must be AI_GATEWAY_API_KEY. Ensure you're running the command from the root of the build-an-ai-app-starter-sep-25 directory.
Other install errors ("module not found", etc.): Delete node_modules and pnpm-lock.yaml, then run pnpm install again. If issues persist, check your Node.js and pnpm versions or seek help.

Next Up: Your First AI Script

Congratulations! Your development environment is set up and ready.

Now you'll write and run code that uses the AI SDK to interact with an LLM and perform a practical task: extracting structured data from text. Time to start building.

Builders Guide to the AI SDK