LangGraph Adapter

@memfork/langgraph is a LangGraph checkpointer backed by MemForks.

Each graph thread maps to a MemForks branch. Checkpoints are committed through MemForks, so graph state can persist across runs, processes, and machines.

Install

npm install @memfork/langgraph @memfork/core @langchain/langgraph

Provision credentials:

memfork init --quick

Basic Usage

import { StateGraph, MessagesAnnotation } from "@langchain/langgraph";
import { createMemForksCheckpointer } from "@memfork/langgraph";

const checkpointer = await createMemForksCheckpointer({
  treeId: process.env.MEMFORK_TREE_ID!,
  signer: process.env.MEMFORK_PRIVATE_KEY!,
  memwal: {
    accountId: process.env.MEMFORK_MEMWAL_ACCOUNT!,
    delegateKey: process.env.MEMFORK_MEMWAL_KEY!,
    serverUrl: process.env.MEMFORK_RELAYER_URL,
  },
});

const app = new StateGraph(MessagesAnnotation)
  .addNode("agent", myAgentNode)
  .addEdge("__start__", "agent")
  .compile({ checkpointer });

await app.invoke(
  { messages: [{ role: "user", content: "Hello" }] },
  { configurable: { thread_id: "abc123" } },
);

By default, thread_id: "abc123" maps to branch thread/abc123.

Custom Thread-To-Branch Mapping

const checkpointer = await createMemForksCheckpointer({
  treeId,
  signer,
  memwal,
  threadToBranch: (threadId) => `agent/${threadId}`,
});

Use stable thread IDs when you want future runs to build on previous graph state.

How It Works

LangGraph event	MemForks behavior
put() checkpoint	Commits serialized checkpoint state to the thread branch.
getTuple()	Recalls checkpoint state for the thread.
putWrites()	Commits pending writes to branch memory.
proposeMerge()	Opens a MemForks merge proposal between thread branches.

Multi-Agent Branches

This pattern lets parallel workers accumulate knowledge independently, then merge their findings into a supervisor branch.

Cross-Agent Merge

const digest = await checkpointer.proposeMerge({
  fromThread: "research-a",
  intoThread: "supervisor",
  resolverId: process.env.MEMFORK_RESOLVER_ID!,
});

The resolver handles reconciliation and finalization.

When To Use LangGraph

Use the LangGraph adapter when your app has:

• multi-step workflows
• parallel worker agents
• resumable graph state
• long-running tasks
• branch-per-thread memory
• checkpoint recovery after process restarts
• governed merges between agents

For simple chat apps, use the Vercel AI SDK adapter.

Reference Example

See LangGraph Research for a multi-agent research pipeline using worker branches and supervisor merges.