Getting started

turbomem gives your LLM agents persistent, semantically-searchable memory that runs entirely inside your Node/Bun process. Type-safe, adapter-based, and dependency-light. No separate memory server required. Local-first by default; pluggable to edge and serverless.

Deploying to Vercel or edge?

Use upstash-vector or pinecone storage on serverless and edge runtimes. See the Edge guide for setup on Cloudflare Workers, Vercel Edge, and Next.js serverless.

Install

bash

npm install turbomem

The default stack (OpenAI embeddings + PGlite storage) works out of the box. PGlite ships as a dependency, no extra setup.

Environment

Set your OpenAI API key for embeddings and fact extraction:

bash

export OPENAI_API_KEY=sk-...

Basic usage

import { TurboMemory } from "turbomem";

const memory = new TurboMemory({
  embeddings: "openai",
  storage: "pglite",
  extraction: { provider: "openai", model: "gpt-4.1-mini" },
  openai: { apiKey: process.env.OPENAI_API_KEY },
  pglite: { dataDir: ".turbomem" },
});

await memory.init();

await memory.add(
  [
    { role: "user", content: "Hey, I love hiking and I'm training for a half marathon this fall." },
    { role: "assistant", content: "Nice — I'll remember your fitness goals." },
  ],
  { userId: "user_123" },
);

const results = await memory.search("What outdoor activities is the user into?", {
  userId: "user_123",
  limit: 5,
});

for (const { memory: m, score } of results) {
  console.log(`[${score.toFixed(3)}] ${m.content}`);
}

await memory.close();

init() is required before any other method. It runs storage migrations and loads embedding models (for local embeddings). Calling other methods before init() throws NotInitialisedError.

Why embedded memory?

Many agent-memory setups rely on a separate service. A Python server, a hosted platform, or a vector database you operate yourself. For a TypeScript app that often means running another process, managing extra infrastructure, and crossing a network boundary for every memory operation.

turbomem is fully embedded:

	turbomem (embedded)	Server-based memory
Runtime	TypeScript, in-process	Separate server / hosted API
Deployment	`npm install`	Run or host a service
Network hop	None (local)	HTTP per call
Storage	PGlite (WASM Postgres)	External vector store
Best for	TS apps, edge, embedding into a product	Multi-language or managed infra

If you need a cross-language managed platform, a dedicated memory service may fit better. If you're shipping a TypeScript app and want memory as a library, that's turbomem.

Next steps

Configuration - embeddings, extraction, scoping
Storage - PGlite, sqlite-vec, Upstash Vector, and Pinecone
Edge - deploy on Workers, Vercel Edge, and Next.js serverless
Architecture - how the pipeline works
Examples - runnable projects in the repo

Deploy on Vercel / Next.js

Next.js on Vercel?
├── Edge runtime → upstash-vector or pinecone (see Edge guide)
├── Node serverless → upstash-vector or pinecone recommended; pglite only with persistent volume
└── Client-side → turbomem/browser with idb://

The vercel-ai-chatbot example switches between PGlite (local dev) and Upstash (production) via TURBOMEM_STORAGE.

Getting started ​

Install ​

Environment ​

Basic usage ​

Why embedded memory? ​

Next steps ​

Deploy on Vercel / Next.js ​