// not a course. it's my work tool.

Everything I know how to do on Claude,
packaged so you can use it too.

robOS is not a course and not a pile of skills scraped off the internet. It's the system I use to do complex work on Claude every day: memory that remembers for months, skills that pick themselves once you ask, and verification that won't let you publish made-up things.

Built by Adrian Ulmeanu, 30+ years in IT and automation. The same tool I use daily, handed to you ready to work.

Get access How it works ↓

"Search brings you an answer and forgets it. A wiki accumulates and grows."

Ady Insights, on the pattern robOS delivers ready to use.

// pick a scenario:

robOS terminal

// the robOS principle

Doesn't forget. Doesn't lie. Doesn't assume.

Doesn't forget

It remembers what you worked on, decided and learned, for months, not just today's conversation.

Doesn't lie

When it can't back up a claim, it doesn't invent one. It tells you straight what it knows and what it doesn't.

Doesn't assume

If it isn't clear what you want, it asks before starting. It won't act on your money without asking, not even on your ads: it shows you what it would change and why, then you decide.

// so it's clear from the start

What robOS actually is

Three things worth knowing before you mislabel it.

// it's not

Not a course

There's nothing to learn before you use it. If you know how to talk to Claude, you know how to use robOS on day one. No lessons, no modules, no homework.

// it's not

Not skills off the internet

Each skill is a tested working machine, with its own logic, declared dependencies and verification. Not a prompt copied from somewhere. The difference between an engine and a photo of an engine.

// it is

It's my actual work tool

The same one I use daily to solve complex work on Claude. I didn't build it to sell it; I sell it because I use it. The difference between a recipe book and a kitchen already fully equipped.

// the problem

Claude remembers. But not the way you need.

Yes, Claude has had memory since March 2026. It writes a summary every 24 hours, saves what it decides to save, and you don't see exactly what was kept and what was left out. You can't separate one workspace from another inside its memory.

If you work across multiple workspaces, you end up repeating the same things in every new conversation. Not because it forgets entirely, but because it remembers without order.

robOS adds order, discipline and verification on top of what Claude already does well.

// what happens without structure

Without persistent memory, you repeat context from scratch in every new conversation. Yesterday you explained everything. Today you start over.

Without methods that learn from your feedback, every result sounds the same. What it writes for you is identical to what it writes for anyone else.

Without verification, you get answers that sound good but don't hold up. You publish them, then find out too late they were invented.

Without workspaces, when you switch clients you lose all the previous context. Tone, decisions, history. Everything has to be rebuilt manually.

// comparison

Without robOS vs With robOS

Same tasks, two ways of working.

Without robOS

With robOS

Claude decides on its own what to keep and what to drop, without asking you

You see everything that's been saved, search through it, export it whenever you want

You lose what you built last month

Everything you've built accumulates, searchable in 2 seconds

Claude answers even when it's not sure about what it's saying

When it doesn't know, it asks

You re-explain each workspace context every time

Each workspace context loads automatically

Copy-paste between 8 open windows

One place for everything

Working method reinvented every time

Ready-made methods that learn from feedback

// what robOS adds

Three things Claude alone doesn't do.

Remembers everything. You never repeat yourself.

Ordinary AI chats start over every time. robOS remembers what you are working on, what you decided and what you learned, day after day. The next day you pick up where you left off.

You continue, you don't restart

Open the terminal and robOS already knows where you left off yesterday. No recaps, no „let me remind you where we were".

Nothing gets lost

Everything you have worked on gathers in one place that is yours, and you find it in two words, whether it was today or three months ago.

No loose ends

If you closed abruptly or a task was left unfinished, it brings it back to you instead of burying it.

You know why you decided

Important choices stay written with their reasoning. Six months later you don't reconstruct from memory, you have them noted.

Plus: a second brain that knows you

You give it your materials (site, profile, a deck) and robOS builds a map of who you are and what you do, the "second brain" idea on top of an AI model popularized by Andrej Karpathy, a model Google Cloud standardized in June 2026 as the Open Knowledge Format. From then on, when it writes about you or your business, it uses your real facts, not guesses. And each workspace has its own brain: what it knows about one client never mixes with another.

Your second brain, seen as a living map. Each dot is a topic; the lines are the links between them.

Drop a file, a link or some text and it gets added to the brain on its own, no command needed

See everything it knows as a living map, not a list you scroll

Delete or clean up what is no longer relevant, from one place

Anti-improvisation verification

"I don't know" is a first-class answer, not a weakness.

Before writing anything important, it checks whether it can back up each claim

If it can't, it says "I don't know, I'll ask", it doesn't invent

When a skill is missing for what you asked, it says so directly. No workaround improvised.

"I don't know" is a first-class answer, not a weakness

robOS terminal

you:

Write an article about 2026 SEO trends with concrete numbers

robOS:

I found verifiable data for 4 out of 6 points. For the other 2 I don't have reliable sources, I'd rather skip them than invent. Want me to continue with the 4, or will you find the sources for the rest?

Multiple workspaces. Zero mess between them.

A workspace is your container: a client, a personal project, a partnership, a product line. Anything you'd want to keep separate. You switch with one command.

All context switches automatically: voice, brand, memory, output

Tomorrow you open the terminal, you're still in the same workspace

The price doesn't grow with the number of workspaces

"which client is active" see who you're working for now

"switch to Flavor Bistro" everything changes automatically

"client root" back to your base space

active client: Flavor Bistro

Voice: warm, personal, host tone

Audience: urban 25-35, foodies

Tasks: recipe article, Instagram post, newsletter

Last session: Tuesday, newsletter draft

active client: DentPro Clinic

Voice: professional, trustworthy, empathetic

Audience: patients 30-55, dental anxiety

Tasks: FAQ article, Google review response

Last session: Monday, competitor audit

One command. All context switches. Nothing to configure manually.

robOS workspace

$ switch to DentPro Clinic

Active workspace: DentPro Clinic

voice: professional, trustworthy, empathetic

audience: patients 30-55

last session: Monday, competitor audit

Workspace memory loaded. 3 open threads.

// robOS asks you too, it doesn't just execute

The best answers are already in your head.

Most tools sit and wait for you to know exactly what you want. robOS also works the other way around: when it matters, it asks you the questions and pulls out what counts. You don't start from a blank page, you answer one prompt at a time, and at the end it's all written down and reused everywhere.

You type: "onboard me"

15 minutes of simple questions, and robOS knows you: who you are, what you do, how you talk. From then on it writes for your real situation, not a generic one.

You type: "brand voice"

It learns how you write, so everything that comes after (articles, posts, emails) sounds like you, not like a robot.

You type: "ideal client"

It asks until it's clear who you sell to and what hurts them. From there, the message starts to land.

You type: "positioning"

It guides you to the angle that sets you apart, when you sound just like every competitor.

You type: "grill me"

It takes a plan apart with hard questions and finds the holes before they cost you money. You come out with the plan stronger, not flattered.

You type: "level up"

A few targeted questions, then it tells you the next step actually worth automating. No guessing on your own.

You don't need to know what to ask. robOS does. You just answer.

// your property

Your data stays on your computer. And stays yours.

Many AI assistants keep everything you discussed in their cloud. When you cancel the subscription, you lose it all. robOS works differently: everything that matters sits in files on your disk, in a format you can read with any editor.

~/robos/context/

Everything is on your disk

Memory, decisions, audits, brand voice, client profiles. Plain text files. Open them with Notepad, Word, any editor. Nothing encoded, nothing locked.

.env

API keys stay with you

They don't pass through any server of mine. Your Claude key, your Zernio key, any service you use - direct between you and that provider.

cp -r robos/ ~/backup/

Leave with everything, anytime

Moving to another computer? Copy the folder. Want an archive? Onto a USB stick. No account to close, no export to request, no cloud to migrate from.

// open source

The code is readable

You see exactly what robOS does, line by line. Nothing hidden behind a logo. If you want to change something, you can.

// a working day