LangChain Development

When to Use This Skill

Core Expertise

LangChain JS/TS is a framework for building LLM applications:

• Unified interface across model providers (OpenAI, Anthropic, Google, etc.)
• Composable chains and agents
• Built-in tool integration
• RAG (Retrieval-Augmented Generation) support
• LangSmith observability integration

Installation

Package Manager Setup

# Core package
npm install langchain
# or
pnpm add langchain
# or
bun add langchain

# Model provider packages (install what you need)
npm install @langchain/openai
npm install @langchain/anthropic
npm install @langchain/google-genai

# Common integrations
npm install @langchain/community  # Community integrations
npm install @langchain/textsplitters  # Document splitting

Chat Models

Basic Usage

import { ChatOpenAI } from "@langchain/openai";
import { ChatAnthropic } from "@langchain/anthropic";
import { HumanMessage, SystemMessage } from "@langchain/core/messages";

// OpenAI
const openai = new ChatOpenAI({
  model: "gpt-4o",
  temperature: 0,
});

// Anthropic
// Use a real, current model id (never an unversioned alias like "claude-haiku"),
// and omit sampling params — Fable-generation models reject temperature/top_p/top_k.
const anthropic = new ChatAnthropic({
  model: "claude-haiku-4-5",
});

// Invoke with messages
const response = await openai.invoke([
  new SystemMessage("You are a helpful assistant."),
  new HumanMessage("Hello!"),
]);

Streaming

const stream = await openai.stream([new HumanMessage("Tell me a story")]);

for await (const chunk of stream) {
  process.stdout.write(chunk.content as string);
}

Structured Output

import { z } from "zod";

const schema = z.object({
  name: z.string().describe("The name"),
  age: z.number().describe("The age"),
});

const structuredLlm = openai.withStructuredOutput(schema);
const result = await structuredLlm.invoke("John is 30 years old");
// { name: "John", age: 30 }

Prompt Templates

Basic Templates

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

const prompt = ChatPromptTemplate.fromMessages([
  ["system", "You are a {role}."],
  ["human", "{input}"],
]);

const formatted = await prompt.invoke({
  role: "helpful assistant",
  input: "Hello!",
});

Few-Shot Prompts

import { FewShotChatMessagePromptTemplate } from "@langchain/core/prompts";

const examples = [
  { input: "2+2", output: "4" },
  { input: "3+3", output: "6" },
];

const fewShotPrompt = new FewShotChatMessagePromptTemplate({
  examplePrompt: ChatPromptTemplate.fromMessages([
    ["human", "{input}"],
    ["ai", "{output}"],
  ]),
  examples,
  inputVariables: ["input"],
});

Chains (LCEL)

Basic Chain

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

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

// Chain with pipe operator
const chain = prompt.pipe(model).pipe(parser);

const result = await chain.invoke({ topic: "programming" });

Parallel Chains

import { RunnableParallel } from "@langchain/core/runnables";

const parallel = RunnableParallel.from({
  joke: jokeChain,
  poem: poemChain,
});

const results = await parallel.invoke({ topic: "cats" });
// { joke: "...", poem: "..." }

Branching

import { RunnableBranch } from "@langchain/core/runnables";

const branch = RunnableBranch.from([
  [(x) => x.type === "math", mathChain],
  [(x) => x.type === "code", codeChain],
  defaultChain, // Fallback
]);

Agentic Optimizations

Quick Reference

Environment Variables

Common Imports

Key Packages

For TypeScript configuration, tool definition and binding, RAG pipelines, and ReAct agents, see REFERENCE.md.