Build your own Life OS with Claude Code: inbox, tasks, people, calendar, bills, reviews, business intelligence, email and Telegram senses, a complete brand pack, creator, media and video studios, growth tools, an agent cockpit, voice and phone. 55 prompts.
What you build
| Part | What it is | What it does |
|---|---|---|
| Memory | Postgres with pgvector: the Life Graph | Every row has an owner and a workspace; an append-only record of what happened |
| Senses | Inbox, email, Telegram, file intake with zones, calendar feeds, n8n | Everything comes in through bound, signed doors; the AI suggests, you file |
| Household Intelligence | Tasks, people, calendar, Bill Map, money, documents, reviews | Warnings before bills and renewals are due, a timed weekly reset, a minimum week |
| Business Intelligence | KPIs, memory search, morning brief, self-knowledge graph | Answers from your own rows with citations; a knowledge base that proposes its own fixes |
| Studios and growth | Complete brand pack, Creator, media, video, notebooks, leads, SEO, radar, ads, sales coaching | Everything drafted and estimated first; nothing is sent, posted or paid without approval |
| Hands | Agent cockpit, org, website studio, lab, assistant, voice and phone | Agents run inside budgets and gates; the phone speaks only your approved FAQ |
Where this comes from
Life OS started as a research question: what does a personal operating system actually need, and why do most of them get abandoned inside a month? The answer came from two directions, a survey of the systems people keep using (weekly reviews, one capture inbox, the PARA and GTD methods, self-hosted tools like a document manager and a budget app) and a second model's take on the same question, merged into one blueprint. The shape of two earlier AI Auto Base packs is inside it too: the Business Intelligence series and Brand Pack Studio.
It also brings in AI Auto Base's own Business Intelligence class series (email senses, a Telegram channel, file intake with zones, n8n as the nervous system, a module host and a self-knowledge graph) and the full Brand Pack Studio, from strategy and logo suite to UI kit, social templates, print, a brand video kit and a folder-structured guidelines export, and it takes inspiration from the work of two builders. Julian Goldie's agent dashboard pack shaped a large part of it: every capability in its fifty install guides, from model routing and agent runners to media, video, research notebooks, leads, SEO, ads, sales coaching, a website studio, an app and game lab, voice and phone, has a prompt here that makes Life OS do it natively. Andy Hafell's content pipeline app, CreatorGrowth, shaped the Creator Studio; if you want Andy's own tools and community, they are here: skool.com/aimate. Both were rebuilt from scratch in this pack's own stack. The prompts contain none of their code or text.
What you build
One app that runs your business life and your home life from the same place. Same brain, two houses: Business Intelligence at work, Household Intelligence at home. It is the control plane on top of your other tools, not a replacement for them. Your bank, calendar provider and document manager stay where they are; Life OS links to them through sockets and keeps the part none of them can: who owns what, how things connect, what needs you today, and a record of what happened that nobody can quietly rewrite.
What it deliberately does not do: it never pays a bill, never moves money, never sends a message, never publishes a video and never starts an agent job on its own. The AI drafts, ranks and proposes; you approve anything that spends money, makes a promise, or leaves the house or the business with your name on it. That rule is written into the database, and the last prompt adds tests that fail the build if any code path gets around it.
Do you need this? Be honest first
If a paper planner and a shared family calendar are working for you, keep them. This is for the person running a small business and a household at the same time who has already tried the apps and watched them go stale, and who is willing to build the system themselves and own it. It is also a long build: fifty-five sessions. If you want something running tonight, pick a smaller pack first.
The research behind it says Life OS systems die from upkeep, not from missing features. The pack ends by telling you to use it for thirty days before you add anything. Take that seriously.
What you need
- The machine. Any laptop to build on. For the finished app, either a small VPS (about $5-10 a month) or a Mac mini with 64 GB of memory or more if you want local models with Ollama.
- The accounts. Claude Code (Pro, Max or Team), and an OpenRouter key on free models while you build. Nothing else is required: every outside service is a socket with a demo twin that answers by itself.
- The time. 30-45 minutes of setup, then about four weeks at a session or two per evening. The README has a week-by-week path.
What is in the zip
life-os-pack/
README.md what you are building, hosting, accounts, the 30-day path, credits
index.html the clickable guide: setup per platform, every prompt with a copy button
COSTS.md every cost, with the date it was read
CREDITS.md where the ideas come from, what was verified and what was not
paths/ Windows (WSL2), Mac (Apple Silicon), Linux (Omarchy), and the VPS route
prompts/ 55 prompts, 00 to 54, pasted into Claude Code one session at a time
assets/cover.png
What was verified and what was not
Verified on 2026-09-25: the guide was driven in a real Chromium (tabs, progress bar, saved progress, copy buttons, opening straight from disk) with all 55 prompt cards; every file was scanned for keys, private addresses and personal data; and four review passes by a second model (Codex) went over the prompts and pages with 49, 51, 62 and 57 findings, all applied except one rejected with the reason recorded.
Not verified: the five brand-pack prompts (33-37) were added after the fourth pass and have not had a second-model review yet; a fifth pass is scheduled and its fixes will ship as an update. The prompts have not been run in order on a fresh machine, the three setup paths have not been run on fresh machines, the app itself has not been built from these prompts, and none of the sockets have been connected to real services. The pack is distributed as is, as a proof of concept, with no guarantee any step works on your setup. When something breaks, paste the full error into Claude Code and ask it to fix it, then post what broke in the community.
Read the prompt before you run it
Before you paste each prompt, hand it to Claude and ask what it is going to build and what it will not touch. Every prompt ends with a proof you can see: a page, a printed line, a row. If the proof does not show up, stop there and fix it before the next session.
