Skip to content

Getting Started

Memory runs in two shapes. Pick yours, then follow one path — after the first step they are the same work: a project, a blueprint, and a client connected.

Self-hosted

Run Memory on infrastructure you control through a single installer — Railway, Docker, Google Cloud or AWS. Your database, your keys, your network.

Install your own instance

Memory Cloud

Use the instance we run and maintain. Nothing to install and nothing to upgrade — sign in and you are a project away from a working graph.

Use Memory Cloud
Self-hosted Memory Cloud
Who runs it You — the containers, database, object storage and identity stack all run in your account Us
First step Run the installer (about 20 minutes, mostly waiting) Sign in with your account
What you need A Railway, Docker or cloud account, and a domain for a production install An invitation from your organization
Best for Data residency, private networks, on-premise CI, full control of upgrades Getting to a working graph the same day, with no infrastructure to own
Upgrades You choose when to move We roll them out

Whichever path you take, the same three things come next:

  1. A project — the workspace that holds your graph, documents, agents and branches. An organization can own several.
  2. A blueprint — a versioned bundle of object types, agents, skills and seed data that installs a working domain model in one action. You do not start from an empty schema.
  3. A client — an MCP share for an AI client, the memory CLI for a terminal, or the REST API for anything else. All three use the same credentials.
  • Deploy on Railway — the full walkthrough of a self-hosted install, every screen explained.
  • Knowledge Graph — what you are actually storing once you are connected.
  • Blueprints — the CLI’s blueprint commands, for installing the same bundle from a script.