Set model behavior
Define a stable assistant role, tone, or policy for every request.
AI Tools · Const
System Prompt
A system prompt provides initial instructions that guide and constrain a model’s behavior. In the current AI SDK, pass it through instructions, alongside prompt or messages.
How it works
A system prompt gives the model initial instructions that shape and constrain its behavior. In the current API, provide those instructions through the top-level instructions option alongside either prompt or messages; system messages embedded in those fields are rejected by default.
Syntax
generateText({ model, instructions, prompt | messages, allowSystemInMessages? })Parameters
| Name | Type | Description |
|---|---|---|
instructions | string | ... | Model instructions that establish system-level behavior. |
prompt | string | A single user-facing input prompt, used alongside system instructions. |
messages | ModelMessage[] | A conversation message list, usable alongside system instructions. |
allowSystemInMessages | boolean | Opt-in switch for accepting system messages embedded in message data. |
Returns
The page documents system prompts and their configuration, but does not define a standalone return value for them.
Key points
- A system prompt establishes the model’s initial behavioral instructions.
- Its purpose is to steer and limit the model’s behavior and output.
- The current top-level API field for system instructions is
instructions. - Existing histories with embedded system messages require explicit opt-in.
Common mistakes
- Do not place a system message inside
promptormessagesunless you explicitly opt into that behavior withallowSystemInMessages: true.
Example: Using instructions for a system prompt
import { generateText } from 'ai';
import { openai } from '@ai-sdk/openai';
const result = await generateText({
model: openai('gpt-4o-mini'),
instructions: 'You are a concise software engineering tutor.',
prompt: 'Explain dependency injection in one paragraph.',
});
console.log(result.text);The request supplies trusted system-level behavior through the top-level instructions option.
When to use it
Combine instructions with a prompt
Add system-level guidance while sending a one-shot user request.
Guide message-based interactions
Run a multi-turn conversation with trusted message history.
Interview questions
QWhat is a system prompt, and how is it configured?
A system prompt is the initial instruction set used to guide and constrain model behavior. In the current API, it is supplied with instructions, which works with both prompt and messages.
Follow-up: What should be used when an existing trusted history contains system messages? Set allowSystemInMessages: true; otherwise those messages are rejected by default.
QWhere should system instructions be placed?
Use the top-level instructions property for system-level behavior. This keeps the system instructions separate from the user-facing prompt or conversation messages.
Follow-up: Why not put the instruction in messages? Embedded system messages are rejected by default and can create injection risks when histories are not trusted.
QWhat happens when a request contains a system message in its message data?
System messages inside prompt or messages are rejected by default. To accept an existing history containing them, enable allowSystemInMessages: true, but only when the messages are trusted.
Follow-up: What is the security concern? A user who can edit the history may inject a system message that overrides or establishes the system prompt.
Related
generateTextstreamTextpromptmessagesModelMessageallowSystemInMessages