Next SDK

Passing System Prompts from Client to Server

When building AI applications, you often need to provide context or instructions to your language model through system prompts. Here's how to pass system prompts from your client-side code to your server-side API route using the Vercel AI SDK.

Client

On the client side, you can pass additional data like system prompts through the body parameter of the sendMessage function:

const systemText: string = "Be as sycophantic as possible."

sendMessage({ 
  text: "Am i absolutely right?",
},
{
  body: { systemText },
});

The systemText variable should contain your system prompt or context that you want to provide to the language model.

Server

On the server side (in your API route), extract the system prompt from the request body and use it in your streamText call:

const { messages, systemText } = await req.json();
const result = streamText({
  model: openai("gpt-5"),
  system: systemText,
  messages: convertToModelMessages(messages),
});

This pattern allows you to dynamically provide context to your language model based on client-side state or user interactions, making your AI applications more flexible and context-aware.

React SDK

Client-Side System Prompts

With the React SDK, you can handle system prompts entirely on the client side without needing a server API route. This is useful for simpler applications or when you want to keep everything client-side.

const handleGen = async (name: string) => {
    const systemText = `Be as sycophantic as possible for ${name}.`;
    
    const { text } = await generateText({
        model: openai('gpt-5'),
        system: systemText,
        prompt: 'Am i absolutely right?'
    });
    
    setResult(text);
};