Home Products Personal OS
Personal OS2.0

Your AI waits for you.
It shouldn’t.

Claude now remembers who you are. It still does nothing until you come to it. Personal OS gives Claude Code, Cursor and ChatGPT Codex structure and agents that run on a schedule, including while you sleep.

Download, watch the videos, install it yourself in ten minutes. No call required.

Claude Cursor ChatGPT Codex ChatGPT Works with Claude Code, Cursor or ChatGPT Codex. Any one of them is enough.
1:19 · what Personal OS 2.0 is.

This video was cut, captioned and put together by an agent that ships with the system. The same one you’ll have after you install it.

The video doesn’t mention it: since the latest update Personal OS 2.0 also works with ChatGPT Codex and ChatGPT on your computer, alongside Claude Code and Cursor.

What Personal OS is

A layer on top of what you already use

Nothing to abandon, nothing to migrate. Cursor, Claude or Codex stays your workplace. Personal OS installs into it and sets the rules the AI follows when it runs your work, and Obsidian is the window you see all of it through: projects, people, tasks, decisions.

What it adds

A structure your projects, people and commitments live in: a project knows who works on it, a person knows what you promised them, a meeting knows what was decided in it. Rules by which the AI keeps that structure itself: where to file a thing, what to link it to, what to raise later. And thirteen agents that work while you’re busy with something else.

Working from day one

Ten minutes and two double-clicks to install, with nothing to move or convert. After that it fills up as you work. It never asks for time of its own.

You’re not buying a template to fill in. You’re buying a finished system: the same Cursor, Claude or Codex you already work in, turned into something that runs your work with you.

What changes in your work

The AI stops asking who everyone is

“Draft the reply to Anna about the contract” just works. Who Anna is, what she was promised last Tuesday, and where the thread stopped all sit next to the task. You stop re-explaining the same background three times a day.

Commitments stop falling through

The radar says it out loud: Anna has been waiting nine days. You’re not the one remembering; the system is. One catch like that a month pays for the purchase.

The morning starts with the picture already drawn

Everything was assembled while you slept: what’s on today, what moved, who’s waiting. Half an hour of getting your bearings turns into two minutes of reading.

Decisions stop living inside threads

Settle something in a chat and it lands in the project file, not in message two hundred that nobody will find again. A month later you can see not just what was decided but why.

You can see what the AI knows about you

Everything the system knows about you is kept on your own machine, always within reach. Open it, correct it, cut what’s wrong, show it to a colleague. You can’t open an AI’s memory that way.

Your work outlives the tool

Switch from Cursor to Claude, cancel a subscription, lose access to a service. What you’ve built stays with you, and the next tool reads it on day one.

Get the system →

$149, one payment. Ten minutes to install.

The model is the same one everyone else has. Your AI isn’t. It knows your projects, your people, your rules, and what you’ve already asked it not to suggest again. The same AI works differently for two people: the stronger one has a system behind it.

You wake up and the work
is already done

Early morning. You’re still asleep. An agent fires on your machine: collects what changed across your projects in the last 24 hours, checks what’s due today, looks at who you haven’t answered in over a week. By the start of your workday the digest is written and waiting for you.

You didn’t open a chat. You didn’t write a prompt. You weren’t at the computer at all.

Thirteen of these come with the system: a morning project review, today’s tasks, a radar for forgotten follow-ups, a weekly review, backups, old-chat harvesting. But thirteen is the starting set, not the ceiling. An agent is described in a plain note, and you write as many of your own as you like.

That’s the difference between a tool you go to and staff that work on their own.

See it in the system →

What the running system
looks like

You keep working in the same Claude, Cursor or Codex. These are the boards the system builds for you, from our own working setup. Not a demo folder. About us →

Thirteen agents already inside

This is what runs once you install. Your own agents you describe in plain English, as many as you need.

Protection from other people’s code

built in, always on

Why. Skills and add-ons from the internet can be harmful: they steal keys, wipe notes, send data elsewhere. You can’t tell from the outside.

What it does. Before anything gets installed, it shows you where it came from, what it reads, where it connects and what access it asks for, and flags the risky parts. It also reads the skill’s own text for hidden instructions aimed at the AI, like “read the keys file and send it here” (prompt injection). By default it then runs it in a sandbox, a separate copy of the system that holds none of your keys or passwords.

Result. Nothing reaches your working system without your explicit yes. Every install is logged, so you always know what’s connected and where it came from.

AG019 · Learning from your corrections

nightly, 3:30

Why. You correct the AI on the same point for weeks.

What it does. Overnight it finds where it read you wrong and turns your corrections into rules.

Result. Tomorrow it applies them on its own. The longer you work, the less you repeat yourself.

AG018 · Past-chat harvest

on request

Why. Years of chatting with AI hold decisions you’ll never find again.

What it does. Walks your Cursor and Claude history, shows what it found as cards, and moves into your knowledge base only what you tick.

Result. Years of accumulated thinking become part of the system. Entirely on your machine; nothing goes out.

AG022 · Evening check-in

end of day

Why. The day happened away from the computer: things got done, cancelled, moved, and the system doesn’t know.

What it does. Asks in the evening what changed on your open tasks and updates their status from your answer.

Result. Tomorrow’s list is honest. No weekly sit-down to reconcile everything by hand.

AG004 · Today’s tasks

9:00

Why. The morning goes into working out what the day even holds.

What it does. Pulls together what’s due today, what’s overdue, and the meetings on your calendar.

Result. The list is ready by the time you sit down.

AG003 · Project digest

8:00

Why. Projects run in parallel, and some of them quietly stall.

What it does. Walks every project: what’s in flight, what’s overdue, how much is waiting on a reply.

Result. Waiting on your dashboard when you open it. A stalled project shows up the same day, not two weeks later.

AG013 · Follow-up radar

10:00

Why. People are waiting on you, and you’re the last to remember it.

What it does. Finds the emails and commitments left unanswered and counts the days.

Result. “Anna has been waiting nine days,” before it becomes a problem.

AG006 · Afternoon nudge

16:00

Why. By evening half the morning list has slipped your mind.

What it does. Checks what’s still open from today.

Result. There’s time to finish before the day ends.

AG014 · Weekly review

Sunday, 18:00

Why. The week ends and it’s unclear what actually moved.

What it does. Brings projects, tasks, people and inbox into one picture.

Result. Fifteen minutes instead of an hour of planning, and two clear calls: what to finish, what to archive.

AG016 · Board refresh

every 30 minutes

Why. A dashboard is useful exactly as long as it’s current.

What it does. Rebuilds every board: the ones that ship and the ones you make yourself.

Result. Open the bookmark and the numbers are current. Nothing to refresh by hand.

AG017 · Nightly snapshot

22:00

Why. Losing what you’ve built stings more than never building it.

What it does. Captures the state of everything on your machine.

Result. Any note can go back to how it stood last week. It never leaves your computer.

AG002 · Inbox cleanup

3:00

Why. The inbox fills up and you stop looking at it.

What it does. Moves handled items to the archive.

Result. In the morning the inbox holds only what’s waiting on your decision.

AG015 · On this day

8:00

Why. A good thought is lost a month after it’s written down.

What it does. Surfaces what you wrote on this date in past years.

Result. Ideas and promises you’d forgotten come back.

AG021 · Video editing

on request

Why. Editing eats an evening and takes a skill you may not have.

What it does. Cuts, stitches, adds captions and titles, drops in inserts, reframes for vertical, compresses to size. Say what you want in plain words.

Result. A finished cut, made on your own machine. This agent edited the video at the top of this page.

Thirteen is what’s included, not what’s required. Switch off what you don’t want, move the times to suit you, rewrite what each one says in plain English: the agent that writes at nine will write at seven if that’s better for you.

Mail and calendar connect if you want them, and there’s a separate guide for Google in the kit. The system works without any connections; the reminders simply stay inside it.

Get the system →

$149, one payment. Ten minutes to install.

Memory is not a system

In 2026 Claude learned to remember: facts about you, your projects, your preferences. It’s genuinely good, and it genuinely works.

But look at what it actually is: a list of topics the AI keeps about you. Here’s where it stops:

  • It remembers you have a project. It doesn’t know the next step or who owns it
  • It remembers your partner’s name. It won’t tell you that you owed him an answer nine days ago
  • It decides what to keep. You can’t open the store and arrange things the way your work actually runs
  • And most of all: it does nothing until you open a chat. Not in the morning, not on Friday evening, not while you’re away

Anthropic’s own documentation says it plainly: memory does not run scheduled tasks or operate as autonomous agents. That’s not a flaw. It just isn’t what memory is for.

Count up last month: how many times did you re-explain your own projects to an AI? How many rules did you set again from scratch? How many promised replies did you remember yourself, a week late?

Memory answers “who are you.” A system answers “what’s happening and what’s next.”

We’re not going to pretend Claude’s memory is bad. It’s good. It’s just a different thing.

CapabilityClaude with memoryPersonal OS
Remembers facts about youyesyes
Projects, people and tasks linked to each otherno, it’s a topic listyes, a linked structure
Runs without you, on a schedulenoyes, agents on a schedule
Tells you about a forgotten follow-upnoyes
You can open and edit the whole storepartly, in settingsyes, all of it
Where the store liveson Anthropic’s serverson your disk
You decide what enters the contextno, the AI decidesyes
Works in Cursor and ChatGPT Codexnoyes
Survives cancelling your subscriptionnoyes

You’ll put months of your own work into this system. It stays with you if you cancel the subscription, switch models, or the provider changes the rules a year from now.

These don’t conflict: Claude’s memory keeps working inside the system. It’s just no longer the only thing you have.

To be precise: a local store doesn’t mean the AI runs offline. When Claude or Cursor read one of your notes, that content goes to the provider, exactly like any message you type in a chat. The difference is that the store belongs to you, and you decide what goes into it.

Ten minutes, two double-clicks

Step 1

Download

Right after payment the archive lands in your inbox: the welcome letter, the PDF guide, the Google connection guide, two videos and the system itself. Where to unpack it is in the guide. It’s one step.

Step 2

Double-click the installer

It checks your machine, installs what Obsidian needs, and registers the scheduled agents (launchd on Mac, Task Scheduler on Windows). Then it prints a plain-language report: which agents are running and at what time, which are off and why.

Step 3

Type “run the first session”

Open the folder in Claude, Cursor or Codex and type that one line. The system walks you through the rest itself.

No call to book, no time to find in your calendar, no waiting on us. If something goes wrong, the PDF has a troubleshooting table, and after that you can email us.

Honest requirements

This is a technical product and you’re installing it yourself. Better to filter you out now than disappoint you after payment.

You need
  • A paid Cursor, Claude or ChatGPT subscription. A standard plan (around $20/month) is enough. With Claude it has to be Claude Code, not the browser version; with ChatGPT it has to be Codex on your computer, not chatgpt.com in a browser
  • Obsidian: free, no account needed
  • Mac or Windows
  • No API key needed. The subscription you already pay for covers it
What not to expect
  • No ready-made email agent. Gmail is connected separately: we show you how and help you set it up
  • Not a cloud service: the system lives on your machine, so it has to be on for scheduled agents to fire
  • Not a mobile app of ours: the system runs on your computer. Claude Code and Cursor do let you reach that machine from their own mobile apps; ChatGPT Codex has no mobile app of that kind

Google connects later, with your own keys, if you want it. A separate guide is included.

Who this is for

Yes
  • You’re a founder, consultant, freelancer, or operator. You’re in Claude or Cursor every day. You’re running more than three projects and they live in your head, not in a system. You’ve tried to get organised before, in notes, in a task tracker, in prompts, and every time it fell apart within a month, because keeping it in order was manual work.
No
  • You open an AI once a week, you want something with no setup at all, or you’re looking for “an AI that makes money for me.” That’s not this product, and we’d rather say so here than after you pay.

Pricing

Personal OS
$149one payment

Paid once, for a system you keep. Set against it: the half-hour of morning bearings it gives back every day, and the commitment you didn’t drop.

  • The whole system: structure, rules, templates, thirteen ready-made agents
  • The installer for Mac and Windows: two double-clicks, ten minutes
  • A welcome letter: what to open first, and in what order
  • A step-by-step PDF with screenshots
  • A separate guide to connecting Google
  • Two videos: a welcome and a walkthrough
  • Email support

One payment, no subscription. Personal licence, one person.

Rolled out to your company

We install the system across your team: a kickoff session, then working sessions shaped to each role. Packages and licence terms are further down.

See packages →
Setup guarantee

We get it running. If it won’t start on a supported Mac or Windows machine, send us the installer report and we’ll work it through with you for as long as it takes. If it can’t be made to run for technical reasons, you get your money back.

Prices in USD. A one-time charge with no subscription behind it, so there’s nothing to cancel. Payment by card through Stripe, from anywhere Stripe serves; we never see your card details. The archive, installer, PDF guide and videos are delivered by email right after payment. Personal licence, one person; the LICENSE file is included in the archive. By buying you agree to our Terms of Service and Privacy Policy. Sold by AI4Live LLC, Jupiter, Florida, United States · hello@ai4live.com.

You won’t be left alone
with this

The installer prints a report at the end: what worked and what didn’t. If something didn’t come up, send us that report. We’ll work through it with you until the system is running.

Iskander Zalyalov Iskander Zalyalovfounder, AI4Live LLC
I want it working on your machine. Someone whose system never started won’t leave a review and won’t tell a friend about it, so I answer these myself and I don’t drop them halfway.
support@ai4live.com

Questions

Memory answers “who are you,” and it does that well. It doesn’t build links between projects, people and tasks, it doesn’t give you direct access to the store, and it doesn’t run anything on a schedule. That last point isn’t our opinion. It’s Anthropic’s own wording: memory does not “run scheduled tasks or operate as autonomous agents.” Personal OS adds exactly that, and Claude’s memory keeps working alongside it.

Nothing. The scheduled agents are scripts on your own machine: they read your files, assemble the digests and rebuild the boards. They never call a model at all, so the question of which model they’d pick, and at which reasoning level, doesn’t arise. Nothing meters overnight, even if the most expensive model at its highest setting is the one selected in your editor.

The model costs you when you ask it something yourself in Cursor, Claude or Codex, in whichever model and at whichever level you chose. Exactly as it did before you installed any of this.

The scripts reach the network in one place only: fetching the weather forecast for the morning summary. That’s free.

You can write your own agent that does call a model. Then it runs on your subscription, and which model it asks is your call.

The installer does the work for you: it checks your machine, sets up what’s needed, registers the schedules, and tells you in plain language what worked and what didn’t. The PDF has a troubleshooting table for the handful of things that can go wrong. If you’re still stuck, write to support@ai4live.com. We’ll stay on it until it runs.

A template is folders and notes. It waits for you to put something in it. This adds a rules layer your AI reads and thirteen agents that fire on their own schedule. A template sits still. This system works while you sleep.

Any of the three works. We build on Cursor, for three reasons.

Limits

In Claude you can hit a ceiling mid-task and wait hours. Cursor has separate limits per model: switch to another one and keep going.

Model choice

Cursor connects anything over API: Anthropic, OpenAI, Chinese models, your own local one. You’re not locked to a single provider.

Privacy Mode

Cursor holds zero-data-retention agreements with all providers: your content isn’t used for training and isn’t stored. One honest caveat: this isn’t offline. Requests still pass through Cursor’s infrastructure, which they document themselves.

If you already live in Claude or in ChatGPT and you’re happy there, stay there. The system works the same.

To us: nothing. To Anthropic, OpenAI or Cursor: whatever the AI reads while doing a task, exactly as in a normal chat, no more. The difference from memory is that there the AI decides what to keep on their servers, while here the store is yours and you fill it. If you have data that should never reach a model, keep it outside the system.

The system stays with you and works with another model. Everything lives on your own machine and isn’t tied to any account or provider. The scheduled agents run on your computer too.

They will change; Claude’s memory is the proof. If something breaks for everyone, we fix it and send the fix to owners.

Yes. The system finds them on your machine and asks what to keep.

No. One payment.

Yes. Since the latest update Personal OS 2.0 supports ChatGPT Codex and ChatGPT on your computer, next to Claude Code and Cursor. The store, the rules and the agents are the same files; Codex reads them the same way. The product videos were recorded before that update, so they don’t mention it.

One requirement stays the same: it has to be Codex on your machine, with a paid ChatGPT plan. chatgpt.com in a browser can’t open a folder of your files.

We don’t ship a mobile app, and you don’t need one. Claude Code and Cursor both have their own mobile apps that reach the machine where your system lives, so you can hand it a task from the street and read the result later. ChatGPT Codex has no mobile app of that kind, so with Codex you work from the computer.

Either way the scheduled agents run on your own machine, which has to be on for them to fire.

You install it for them with the same installer. The licence covers the company, not a number of seats, so there’s nothing more to pay. Want us to do it? That’s a separate conversation.

When the system runs
across the whole company

The knowledge stays when the person leaves

Normally context walks out with the employee: what was decided, why, who the client was, where the right version lives. Here it sits in the shared base and stays there.

The base builds itself while people work

Nobody maintains a wiki. Documents and commitments land where the work happens, because each person’s AI writes into the shared structure.

One person’s AI can see what everyone else did

Someone in sales asks their system and gets back the document legal filed yesterday. Not “ask Kate for the file,” but the answer itself.

A new hire gets up to speed without borrowing anyone’s time

During the rollout we build them an agent of their own: it meets the new person, walks them through the company’s projects, people and rules, and answers the questions they’d otherwise ask a colleague. After that you edit it yourselves; we show you how on the session. Instead of two weeks of tapping busy people on the shoulder, they ask a system that holds the company’s history.

A mistake gets corrected once, for everyone

A rule one person arrives at becomes the rule for all of them. Over time the company ends up with one way of working with AI instead of a private version per desk.

The data stays with the company

The base sits on your own server. Not inside someone else’s service that won’t give it back.

The shared base lives on your server with standard access permissions. Obsidian doesn’t resolve two people editing the same note at once, so the setup puts people in their own areas and has them meet in shared ones. We configure that during the rollout. Whatever the AI reads still goes to the model provider: a shared server doesn’t make the data private, it makes it yours.

Obsidian itself is free for business use. No seats, no annual renewal: you pay nothing for the software.

We work in groups of up to six: that’s the size where everyone still gets to ask a question and share their screen. One kickoff session for everybody, then two working sessions per group.

PackagePeopleSessionsWhat’s includedPrice
Launchup to 53A kickoff session and two working ones. Each person’s setup shaped to their role. We build agents around your processes together with you. 30 days of support.$5,000
Departmentup to 105A kickoff session and two for each of two groups. Each person’s setup shaped to their role. Agents around your processes. 60 days of support.$9,000
Companyup to 187A kickoff session and two for each of three groups. A shared knowledge base on your server with access separation. Training for an in-house administrator so you add agents without us. 90 days of support.$18,000

Above eighteen people, each further group of up to six is two more sessions and $3,000 on the package.

After that the company rolls the system out to new hires itself, free and without limits. We charge for our time: the sessions, the setup shaped to each role, the first agents built around your processes, and training whoever carries it on without us.

A company licence, not a seat count

One payment, no subscription. Install it for everyone who needs it, today and a year from now when new people arrive. Bought the five-person package and grown to twelve? Roll out the rest yourselves, at no extra cost.

We charge for our time on the rollout, not for copies.

The one limit: the system can’t be resold or built into a product you charge for. Inside the company, do as you like: your own agents, your own boards, your own rules.

The system, the installer, the guide and two videos. One payment, no subscription. If it doesn’t run, write to support and we’ll fix it.

Your AI already remembers.
Now give it work