@langgraphjs/toolkit

Core toolkit for LangGraph.js — agent templates, middleware, and utilities not included in @langchain/langgraph.

Install

npm install @langchain/langgraph @langchain/core @langchain/openai @langgraphjs/toolkit

What It Provides

import {
  // Agent templates — domain-specific presets not in @langchain/langgraph
  createResearchAgent,
  createCodingAgent,
  createConversationalAgent,
  createDataAnalysisAgent,

  // Long-term memory — conversation summarization beyond MemorySaver
  AgentMemory,

  // Token management — cost tracking and circuit-breaking
  TokenBudget,

  // Production middleware — not in @langchain/langgraph
  RateLimiter,
  withRetry,
  StructuredLogger,
} from "@langgraphjs/toolkit";

These are not available in @langchain/langgraph or @langchain/langgraph/prebuilt.

Quick Start

import { ChatOpenAI } from "@langchain/openai";
import { createResearchAgent, TokenBudget } from "@langgraphjs/toolkit";

const budget = new TokenBudget({ maxTokensPerRequest: 50_000 });

const agent = createResearchAgent({
  llm: new ChatOpenAI({
    modelName: "gpt-4o",
    callbacks: [budget.createCallbackHandler()],
  }),
});

const result = await agent.invoke({
  messages: [{ role: "human", content: "Research LangGraph TypeScript" }],
});

console.log("Tokens used:", budget.getUsage().totalTokens);

Agent Templates

| Template | Use Case | |----------|----------| | createResearchAgent | Web search, fact verification, report synthesis | | createCodingAgent | Code analysis, generation, refactoring | | createConversationalAgent | Multi-turn dialogue with context retention | | createDataAnalysisAgent | Data processing, statistics, insights |

AgentMemory

Long-term conversational memory with summarization. Unlike MemorySaver (which stores raw message history), AgentMemory compresses conversations into summaries and extracts entities.

import { AgentMemory } from "@langgraphjs/toolkit";

const memory = new AgentMemory({ llm: model, maxMessages: 20 });
await memory.update(messages);
const context = memory.getContext(); // Compressed summary

TokenBudget

Track and limit LLM token usage with automatic circuit-breaking.

import { TokenBudget } from "@langgraphjs/toolkit";

const budget = new TokenBudget({
  maxTokensPerRequest: 50_000,
  maxTokensPerSession: 500_000,
  onBudgetExceeded: (usage) => console.warn("Budget exceeded:", usage),
});

Middleware

import { RateLimiter, withRetry, StructuredLogger } from "@langgraphjs/toolkit";

// Rate limiting for LLM API calls
const limiter = new RateLimiter({ maxRequests: 10, windowMs: 60_000 });

// Exponential backoff retry
const result = await withRetry(() => agent.invoke(input), { maxRetries: 3 });

// Structured logging for agent execution
const logger = new StructuredLogger((entry) => console.log(entry));

Documentation

Full guides and tutorials at langgraphjs.guide.

Telemetry

This package runs a lightweight environment compatibility check on install and reports anonymous diagnostics (platform, Node version, architecture). No source code, tokens, or credentials are transmitted. Respects DO_NOT_TRACK.

LANGGRAPHJS_TELEMETRY_DISABLED=1 npm install

License

MIT