Stop Juggling API Keys: Setting up Vercel AI Gateway

I remember spending an entire afternoon trying to get three different LLM providers to play nice together. I had to manage separate rate limits, different auth headers, and varying JSON structures. It felt like I was writing boilerplate instead of building features. When I found out OpenClaw supports the Vercel AI Gateway, I switched immediately. It lets me hit hundreds of models through one API key and a single endpoint.

Using the gateway means I don’t have to update my code every time I want to try a new model. I just change a string in my config and keep moving.

What You’ll Need

• An active Vercel account with an AI Gateway set up.
• The OpenClaw CLI installed on your machine.
• Your AI_GATEWAY_API_KEY.
• A terminal window open.

Quick Start

You can get this running in about two minutes. First, you need to tell OpenClaw to use the gateway for authentication.

1. Run the onboarding command:

openclaw onboard --auth-choice ai-gateway-api-key

2. Open your configuration file and set your primary model. Use the vercel-ai-gateway prefix followed by the provider and model name:

{
  agents: {
    defaults: {
      model: { primary: "vercel-ai-gateway/anthropic/claude-opus-4.6" },
    },
  },
}

If you prefer to do everything in one command without the interactive prompts, use this:

openclaw onboard --non-interactive \
  --mode local \
  --auth-choice ai-gateway-api-key \
  --ai-gateway-api-key "$AI_GATEWAY_API_KEY"

Troubleshooting

Sometimes things don’t work on the first try. Here are the two most common hiccups:

• API Key is missing in the background: If you run OpenClaw as a daemon (using launchd or systemd), the process might not see your shell’s environment variables. Make sure your AI_GATEWAY_API_KEY is saved in ~/.openclaw/.env so the daemon can find it.
• Incorrect Provider String: The gateway uses an Anthropic-compatible API. Double-check that your model path follows the provider/model-name format exactly as shown in the Vercel dashboard.

If you hit a wall that isn’t covered here, you can always ask the AI Setup Assistant for a hand.

What’s Next

• Model Configuration Guide
• Advanced Onboarding Options