RetainDB
SDK

Browse docs

SDK

Tap to expand

Contribute

SDKUpdated 2026-03-18

LangGraph Adapter

Store LangGraph checkpoint state in RetainDB so stateful agents can resume by thread.

Applies to: @retaindb/sdk

For direct agent memory, use `db.user(userId).runTurn(...)` from the main `RetainDB` class — it handles retrieve, inject, generate, and store automatically.

The LangGraph adapter is for teams already using LangGraph's checkpointer pattern for agent state.

Installation

bash
npm install @retaindb/sdk

Usage

typescript
import { RetainDBClient, createLangGraphCheckpointAdapter } from "@retaindb/sdk";
import { createReactAgent } from "@langchain/langgraph/prebuilt";
import { ChatOpenAI } from "@langchain/openai";

const client = RetainDBClient.fromEnv(); // reads RETAINDB_API_KEY

const checkpointer = createLangGraphCheckpointAdapter(client, {
  userIdPrefix: "langgraph-thread",
  defaultCheckpointNs: "default",
});

const agent = createReactAgent({
  llm: new ChatOpenAI({ temperature: 0 }),
  tools,
  checkpointer,
});

const config = { configurable: { thread_id: "conversation-1" } };

await agent.invoke({ messages: ["Hi"] }, config);
await agent.invoke({ messages: ["What did I say?"] }, config);

Adapter options

OptionTypeDefaultDescription
userIdPrefixstring"langgraph-thread"Namespace for checkpoint storage
defaultCheckpointNsstring"default"Default checkpoint namespace

The adapter derives user_id and session_id internally from the thread config. You do not pass them directly.

Why userIdPrefix exists

Checkpoints are stored as RetainDB memories. userIdPrefix namespaces those writes so LangGraph state doesn't mix with your application's user memories.

Next step

Was this page helpful?

Your feedback helps us prioritize docs improvements weekly.