🎞️ LangGraph Cassette

Record once. Replay forever.
Build and debug LangGraph / LangChain agents without burning tokens.

🚀 Why

Calling real LLMs for every test or iteration is slow and expensive.
LangGraph Cassette lets you:

• Record real API responses once
• Re-run them offline, deterministically
• Inspect token usage and costs per call

✨ Features

• 🔄 Record → Replay OpenAI (or any LangChain model) calls
• 💾 JSON cassettes for easy diffing and version control
• 💰 Token & Cost logger
• 🧩 Mock LLM for offline prototyping
• ⚙️ Modes: live, record, replay, auto
• 🪄 LangGraph-first ergonomics

📦 Installation

npm install langgraph-cassette

Usage in code

import CassetteLLM from "langgraph-cassette/core/index.js";
import { createTokenLogger } from "langgraph-cassette/utils/tokenLogger.js";

const logger = createTokenLogger();
const llm = new CassetteLLM({
  mode: process.env.CASSETTE_MODE || "auto",
  cassetteDir: ".cassettes",
  modelOptions: { model: "gpt-4o-mini", temperature: 0 },
  logger,
  verbose: true,
});

const messages = [{ role: "user", content: "Explain LangGraph Cassette." }];
const ai = await llm.invoke(messages);

console.log(ai.content);
console.log("Session summary:", logger.summary());

Raise Issues

https://github.com/arijit1/langgraph-cassette/issues

Example

Record Once

CASSETTE_MODE=record CASSETTE_DIR=.cassettes
node examples/langgraph.mjs

Replay offline

CASSETTE_MODE=replay CASSETTE_DIR=.cassettes node examples/langgraph.mjs

TERMINAL COMMANDS

CASSETTE_MODE=record CASSETTE_DIR=.cassettes node examples/langchain.mjs

or

CASSETTE_MODE=replay CASSETTE_DIR=.cassettes node examples/langchain.mjs

or

CASSETTE_REPLAY_MISS=mock CASSETTE_MODE=replay node examples/langchain.mjs

or

CASSETTE_REPLAY_MISS=live CASSETTE_MODE=replay node examples/langchain.mjs

or

CASSETTE_MODE=record CASSETTE_DIR=.cassettes node examples/langgraph.mjs