Skip to main content
AtomicMemory is a memory layer you embed where your AI code already runs. It captures context, grounds model generations in prior interactions, and carries knowledge across sessions — from a direct SDK call, a CLI, an MCP server, a framework adapter, or a host plugin. Start local, move to hosted, or self-host on your own infrastructure — the choice can change later without rewriting your application.

Quickstart

Install AtomicMemory and make your first memory call in minutes

Core Concepts

Understand memory scopes, ingestion, retrieval, and claim versioning

SDK Reference

Explore the TypeScript SDK — MemoryClient, providers, and storage

API Reference

Full HTTP API reference for the AtomicMemory Core service

​
Choose your integration path

AtomicMemory fits into however you build AI applications. Pick the path that matches your stack.

TypeScript SDK

Direct SDK integration for Node.js applications

CLI

Manage memory from the command line or shell scripts

MCP Server

Expose memory tools to any MCP-compatible agent host

LangChain

Adapter for LangChain JS agents and LCEL chains

Vercel AI

Composable primitives for the Vercel AI SDK

Claude Code

Persistent memory across Claude Code sessions

​
How it works

1

Run AtomicMemory Core

Start a local or hosted Core instance — a single Docker command gets you running in seconds.
2

Install the SDK or adapter

Add @atomicmemory/sdk to your project, or pick a framework adapter for LangChain, LangGraph, Vercel AI, Mastra, or OpenAI Agents.
3

Ingest context

After each model turn, call memory.ingest() to persist the conversation as searchable semantic memories scoped to the user, agent, or thread.
4

Retrieve and ground

Before each model call, search for relevant prior memories and prepend them as context — giving your agent accurate recall without growing context windows.
AtomicMemory is local-first where supported and hosted where convenient. You can start with a local Docker instance and switch to a hosted service later without changing your application code.