Skip to content

Get started

Applies to OML v0.6.x · Status: Stable · Last verified 2026-11-06

Every client needs two settings: the server's address and a key. The server derives everything else from the key.

You are You have Open
Trying OML PostgreSQL or Docker Quickstart
Deciding whether to adopt OML nothing Evaluate OML
An agent user an agent key (get one) Install the CLI, then Claude Code, claude.ai and your phone or another tool
Running a server a machine with Docker, or a PostgreSQL host Run your own server
An administrator an admin key Administer a server

Something failed? Run oml doctor, then see troubleshoot the client or the FAQ.

Sign up on the public server: no key yet

Once the operator opens sign-up on the public server (ADR-0068, proposed), you need only a Google or GitHub account. Sign in at https://memory.doublegate.io/web, accept the terms, and you land on /web/start:

  1. Choose your client: claude.ai, Claude Desktop or the phone app; Claude Code; or another MCP tool.
  2. One button issues the right credential, shown once: a connect code and the connector URL for claude.ai, a key and the plugin's install commands for Claude Code, or the URL and a key for the rest.
  3. The page waits for your agent's first connection, then for its first memory, which you review on the same site.

You get a private space nobody else can read, up to 1,000 memories, 5 keys and 5 connections; the operator can raise a limit. Your agents' writes wait for your review. The operator of the server can read what you store. Export everything as JSON, or ask for your account to be deleted, from /web/start.

Integrations at a glance

Tool How it connects You get Page
Claude Code plugin over MCP, installed by oml setup claude-code the three tools, a session-start hook, a librarian agent Connect Claude Code
claude.ai, Claude Desktop connectors, Claude phone apps custom connector with a one-time connect code the three tools claude.ai and your phone
Codex CLI, Cursor, Gemini CLI, VS Code, Claude Desktop files from the agent kit the three tools over MCP Other agent tools
Your own code Python and TypeScript SDKs write, read and review calls Client SDKs
A reviewer crew HTTP review loop a crew that approves or rejects Reviewer crews