Skip to content
Me docs

Me — user guide

Me is an always-on assistant platform for one person, a household or a small company. It keeps a durable model of your world (messages, tasks, notes, people, calendars, money, AI sessions) on a server you run, and it exposes all of it through one command vocabulary: the me command line, and the same commands as MCP tools for AI agents such as Claude Code.

You don’t log into a dozen web apps to read, triage, act or automate. Your AI assistant doesn’t need to learn twenty vendor APIs. Both of you use the same commands, under the same policy, with the same audit trail.

Status: pre-production. Me runs today as a local install on a Windows PC. Everything in this guide was checked against main. Anything still in an open pull request is marked coming soon, with a link to the PR.

This guide will be published at docs.molga.work, with search (coming soon: PR #162). Until then, read it here on GitHub, in docs/guide/.

Who it is for

Section titled “Who it is for”
You are What Me gives you
A person who lives in an AI tool (Claude Code) One place for tasks, notes, people, mail, chats, calendar and bank statements, which your AI operates for you within limits you set
A household A shared tenant: bills, chores, recurring tasks, documents
A small-business operator Workflows, delegation to people and roles, approvals, audit
An AI agent Typed commands with clear outcomes: done, waiting for approval, denied, or “may have happened, check first”

What makes it different

Section titled “What makes it different”
  • It owns the “in-between” data. No single vendor holds the links between this email, that invoice, this person and that AI session, or your own notes, tags, follow-ups and decisions. Me does.
  • Agents act with delegated authority, never as you. An AI gets an agent identity and a delegation: which operations it may run, up to which risk and sensitivity, until when. Riskier operations wait for your approval; more sensitive data is refused. A payment needs your passkey, always.
  • People get conversations, not scripts. The system composes messages to people from intents (purpose, facts, what it needs to know) and reads their free-text replies back into structured answers.
  • Integrations are real. Microsoft 365, Teams, Telnyx, Cloudflare and Claude connect as live accounts, not simulators. Fixed wording is used only as a fallback: when no model is connected, or when a message is more sensitive than the model or channel is allowed to carry (see Sensitivity).
Section titled “Read next”
  1. Concepts: tenants, principals and agents, integrations, attention, triggers, workflows, sessions and approvals.
  2. Getting started: install on a PC, log in, enable plugins, create your first task and note, connect Claude Code and your first integration.
  3. Integrations: Microsoft 365 calendar, Teams and mail; Telnyx SMS and voicemail; Cloudflare mail; Claude; devices; and Google calendar and Gmail, which are coming soon (PR #120, PR #134).
  4. Operator guide: install.ps1, the Cloudflare tunnel, backups and configuration keys.
  5. CLI reference: every command and flag, generated from the live command catalog.

Other sources:

  • me help and me help <command…> always describe the server you are connected to. They are the source of truth.
  • docs/DEMO.md: flows verified end to end on a live install, with the output you should see.
  • The me skill: the guide your AI agent reads (me skill install).
  • The specification: the full design, including parts that are not built yet.