Skip to main content

​
Overview

Prompt templates help you create reusable, parameterized prompts for language models.

​
ChatPromptTemplate

Template for chat models with message roles. Import: 
import { ChatPromptTemplate } from "@langchain/core/prompts";

​
Basic Usage

const prompt = ChatPromptTemplate.fromTemplate(
  "Tell me a joke about {topic}"
);

const formatted = await prompt.invoke({ topic: "cats" });
// Returns: [HumanMessage("Tell me a joke about cats")]

​
Multiple Messages

const prompt = ChatPromptTemplate.fromMessages([
  ["system", "You are a helpful assistant that speaks like a {character}"],
  ["human", "{input}"]
]);

const messages = await prompt.invoke({
  character: "pirate",
  input: "What is the capital of France?"
});

​
MessagesPlaceholder

Insert dynamic message arrays: 
import { MessagesPlaceholder } from "@langchain/core/prompts";

const prompt = ChatPromptTemplate.fromMessages([
  ["system", "You are a helpful assistant"],
  new MessagesPlaceholder("history"),
  ["human", "{input}"]
]);

const messages = await prompt.invoke({
  history: [
    new HumanMessage("Hi, my name is Alice"),
    new AIMessage("Hello Alice!")
  ],
  input: "What's my name?"
});

​
PromptTemplate

Template for completion models (text-in, text-out). 
import { PromptTemplate } from "@langchain/core/prompts";

const prompt = PromptTemplate.fromTemplate(
  "Write a {length} poem about {subject}"
);

const formatted = await prompt.invoke({
  length: "short",
  subject: "the ocean"
});
// "Write a short poem about the ocean"

​
Partial Variables

Set some variables upfront: 
const prompt = ChatPromptTemplate.fromTemplate(
  "Tell me about {topic} in {language}"
);

const partialPrompt = await prompt.partial({
  language: "Spanish"
});

// Now only need to provide topic
const messages = await partialPrompt.invoke({ topic: "AI" });

​
Few-Shot Prompts

Include examples in your prompt: 
import { FewShotPromptTemplate } from "@langchain/core/prompts";

const examples = [
  { input: "happy", output: "sad" },
  { input: "tall", output: "short" },
  { input: "hot", output: "cold" }
];

const examplePrompt = PromptTemplate.fromTemplate(
  "Input: {input}\nOutput: {output}"
);

const fewShotPrompt = new FewShotPromptTemplate({
  examples,
  examplePrompt,
  prefix: "Give the antonym of each word:",
  suffix: "Input: {input}\nOutput:",
  inputVariables: ["input"]
});

const result = await fewShotPrompt.invoke({ input: "big" });

​
Prompt Composition

Combine prompts: 
const systemPrompt = ChatPromptTemplate.fromMessages([
  ["system", "You are a {role}"]
]);

const userPrompt = ChatPromptTemplate.fromMessages([
  ["human", "{input}"]
]);

const combined = systemPrompt.pipe(userPrompt);

​
Using with Models

import { ChatOpenAI } from "@langchain/openai";
import { ChatPromptTemplate } from "@langchain/core/prompts";

const model = new ChatOpenAI();
const prompt = ChatPromptTemplate.fromTemplate(
  "Tell me a {adjective} joke about {topic}"
);

const chain = prompt.pipe(model);

const response = await chain.invoke({
  adjective: "funny",
  topic: "programmers"
});

​
Prompt Methods

​
invoke
method
Format the prompt with variables
async invoke(values: Record<string, any>): Promise<BaseMessage[]>
​
format
method
Format as a string (for PromptTemplate)
async format(values: Record<string, any>): Promise<string>
​
partial
method
Create a new prompt with some variables pre-filled
async partial(values: Record<string, any>): Promise<PromptTemplate>

​
Template Syntax

​
Variable Substitution

const prompt = ChatPromptTemplate.fromTemplate(
  "Hello {name}, you are {age} years old"
);

​
Conditional Content

const prompt = ChatPromptTemplate.fromTemplate(
  "User info: {name}{#if premium} (Premium Member){/if}"
);

​
Prompt from File

import { readFileSync } from "fs";
import { ChatPromptTemplate } from "@langchain/core/prompts";

const template = readFileSync("./prompts/assistant.txt", "utf-8");
const prompt = ChatPromptTemplate.fromTemplate(template);

​
Best Practices

Choose variable names that clearly indicate what value should be provided:
// Good
"Summarize the following {article_text}"

// Avoid
"Summarize the following {text}"
Put behavior instructions in system messages:
ChatPromptTemplate.fromMessages([
  ["system", "You are a helpful assistant. Always respond in {language}"],
  ["human", "{input}"]
])
This keeps your prompt flexible:
new MessagesPlaceholder("chat_history")

Prompt Engineering Guide

Learn prompt engineering best practices

Chat Models

Using prompts with chat models

Build docs developers (and LLMs) love

Get started for freeTalk to us