← All docs

// Docs

Getting started.

Klyr is a Claude Code plugin that puts your DNA into Claude, so it stops being a generic assistant and becomes your clone. This page takes you from the invitation email to your first conversation: every click, every command, every question explained. If you've never opened a terminal, you're exactly who this was written for.

No invite yet? Klyr is invitation-only for now. Join the waitlist and you'll be first when a spot opens. No promises.

Part one: install it

One command does the heavy lifting.

What you need

• Your Klyr email address. The address that received your Klyr welcome or invitation email: it's the one your access is tied to, and you'll sign in with it once along the way. Nothing to download, no key file to keep.
• A Claude subscription (Pro or Max). Klyr thinks on the Claude subscription you already have; you'll sign in to it along the way.
• A Mac, a Windows PC, or a Linux machine. On Windows, plain Windows is enough: no WSL, no developer tools, nothing extra to install first. That covers Hosted memory, which is what nearly everyone runs; Local memory on Windows is the one exception, and Question 1 below says so plainly.
• About 15 minutes. Most of it is the installer working while you watch.

Step 1: open a terminal

A terminal is just a plain window where you type an instruction and press Enter. That's the whole skill.

• On macOS: press Cmd + Space, type Terminal, and press Enter. A window opens with a blinking cursor. That's it, you're in a terminal.
• On Windows: open the Start menu, type PowerShell, then right-click it and choose Run as administrator and click Yes on the prompt. (Windows asks this so the installer can set up Git for you.) You'll see the same blinking cursor.
• On Linux: on most systems Ctrl + Alt + T, or find Terminal in your applications.

Step 2: paste the install command

Copy the command for your machine, paste it into the terminal, and press Enter.

curl -fsSL https://getklyr.ai/install.sh | bash

irm https://getklyr.ai/install.ps1 | iex

When your browser opens, sign in with the email that received your Klyr welcome or invitation email. You'll get an email with a link; click it, and the installer carries on by itself. After that, a few minutes of text scrolling by is normal — it may ask for your computer's password (nothing appears while you type a password; that's a terminal security habit, not a bug). Wait for the message saying setup is done, and leave the window open.

What the installer sets up, so nothing here is a mystery: git (a standard tool for fetching code), the Claude Code app, access to the Klyr code, the Klyr plugin itself, and a small helper called uv that Klyr uses to run. All standard, widely used tools.

Interrupted, or ended with an error? Just run the same command again: the installer is safe to re-run and picks up where it left off. One exception, and only if you deliberately moved off the default release track: the command above puts this machine on stable, so if you are on beta or canary, re-run your channel's command instead. Still stuck? Write to [email protected] and say where it stopped.

Step 3: open Claude Code for the first time

The installer set up Claude Code, the app your clone lives in. It runs inside the terminal: no icon appears anywhere, and that's normal. To start it, open a terminal (same as before: Cmd + Space then Terminal on a Mac, Start menu then PowerShell on Windows) and type:

claude

Since this is your very first launch, Claude Code introduces itself:

1. It may ask you to pick a color theme. Use the arrow keys to choose, press Enter. Any theme is fine.
2. It asks you to sign in to Claude. Your browser opens; log in with your Claude account (the one with the Pro or Max subscription). This is your Claude account, not necessarily the email that received your Klyr invitation; that one comes later.
3. Back in the terminal, you'll see a chat: you type at the bottom, Claude answers above. That's Claude Code. You never need to learn more of the terminal than this.

Part two: meet /klyr:init, a guided setup that talks to you

Step 4: start the guided setup

In Claude Code, type this and press Enter:

/klyr:init

Commands that start with / are how you ask Claude Code to do something specific. This one walks you through Klyr's setup, one simple question at a time, in plain language. A few things to know before it starts:

• Questions appear as a small menu. Move with the arrow keys or press the option's number, then Enter. You can always type your own answer instead of picking one.
• Most questions put the sensible default first. A few deliberately don't recommend anything: those are real choices that are genuinely yours (the memory question below is one). This page tells you what to expect either way.
• Nothing is final and nothing breaks. /klyr:init is safe to run again anytime: it remembers where you were and continues, it never redoes or erases what's done.

Here is every question it will ask, in the order you'll meet them.

Question 1: where should your clone's memory live, Hosted or Local?

Your clone remembers everything you tell it. This question decides where that memory physically runs. It's the one big choice of the setup, and there's no recommended tag on it because both answers are legitimate:

• Hosted: your memory runs in your own private space on Klyr's servers. This is the paid tier: it works on any machine (even a modest laptop), follows you across devices, needs nothing installed or maintained, and runs on fast datacenter hardware. You sign in with the email that received your Klyr welcome or invitation email.
• Local: everything runs on this one computer, free, and nothing ever leaves it. The trade-off: it needs a strong machine (16 GB of memory is the floor, plus around 23 GB of free disk for the on-device AI models), some background software, and your memory stays tied to this computer. Klyr checks your machine first and tells you honestly if it can't handle self-hosting. On Windows, Local needs WSL2 plus Docker Desktop — the memory engine runs in Linux containers, and plain Windows has nowhere to put them. Mac and Linux need neither. So on a plain Windows PC, Hosted is the answer.

If you pick Hosted, here's the little dance that follows. Claude Code only connects to new memory servers when it starts up, so the setup pauses once for that:

1. When /klyr:init tells you to, close Claude Code and open it again: type /exit, then type claude to start it back up.
2. Run /mcp, pick klyr, and sign in in the browser. Use the email that received your Klyr welcome or invitation email: that's the address your access is tied to. You'll receive an email with a link; click it, and you're in. Nothing to copy or paste, and the link expires quickly by design.
3. Run /klyr:init again. It knows exactly where you left off and carries on.
4. Already use Klyr on another machine? It will notice and offer to bring your existing memory onto this one instead of starting from scratch. Take that offer.

How to choose: pick Hosted if you want zero maintenance and your clone on every machine you use. Pick Local if you'd rather keep everything on one strong computer for free, and you're on a Mac, on Linux, or on Windows with WSL2 and Docker Desktop already set up. Either answer is legitimate, and you can change your mind later.

Question 2: the interview, letting it get to know you

Next, your clone offers to interview you. Klyr calls it KYH, Know Your Human. It asks gentle questions about what you're working on, how you decide, what matters to you, and writes what it learns into its memory. You just talk; it does the writing.

• It's skippable: you can say "later" and set up the rest first.
• It's resumable: stop mid-interview whenever you want; it continues where you left off, and never re-asks what it already knows.
• It's repeatable, and should be. Run /klyr:kyh again regularly: your goals move, and each session makes the clone sharper about who you are now.

Have a past with ChatGPT or Claude? You can import your chat history so your clone starts already knowing you. Do it, but know what it is: your history is where you've been; the interview captures where you're going. An import jumpstarts memory, it doesn't replace KYH.

Recommended: do at least a short first interview now, import your history if you have one, and make /klyr:kyh a regular habit.

Question 3: naming your clone

Your clone gets a name: it's what you'll call it, and what it signs its identity with. Pick anything you like, or ask it to name itself; it will happily propose one. (Everything technical about how your memory is filed happens silently. No question about it will ever reach you.)

Recommended: don't overthink it. Pick something short, or let it choose. You can rename it later.

Question 4: how it behaves with you

By default, your clone is set up as a sparring partner: it challenges your assumptions, pushes back when something doesn't add up, and tells you what it actually thinks instead of agreeing with everything. That's the setting most people end up wanting; a clone that only nods is just a mirror.

Recommended: keep the sparring-partner default. If it's ever too much or too little, just tell it. The behavior is yours to change, anytime.

Question 5: your vault, where your clone keeps what you share

A vault is nothing exotic: it's a plain folder on your disk. The whole idea is one simple loop: you share, your clone files it there. You never write a special format, you never reorganize anything, and your working documents stay wherever they live today: Google Docs, Notion, anywhere. What you share with your clone in conversation is what lands in the vault.

You may already keep a notes folder, an Obsidian folder, any folder of text files, and it can live anywhere on your disk. Klyr never assumes or goes hunting: it asks. If you have one, give its location and choose "keep my folder exactly as it is": Klyr works inside your layout without adding or moving a thing. If you don't have one, let Klyr create a fresh folder where it suggests.

Left to its own devices, Klyr files what you share using PARA: an inbox for quick capture, then Projects, Areas, Resources, Archives. That's the clone's filing habit, not homework for you: it's entirely optional, and saying no changes nothing about what Klyr remembers. The curious can read the longer story on the blog.

Recommended: already keep notes in a folder? Point Klyr at it and keep your layout as it is. Starting fresh? Accept the suggested location and let your clone handle the filing.

Question 6: the Klyr look for your terminal

Optional and purely cosmetic: a warm color theme, plus a one-line status bar at the bottom of Claude Code that keeps you oriented. The status line looks like this:

Nova (Opus) · my-notes main · 247 ✦ hosted · 34% ctx

Reading left to right:

• Your clone's name and the model it's currently thinking with.
• The folder you're in, and the git branch if there is one.
• How many conversations it has banked into memory (the ✦ number) and which memory mode you're on, hosted or local.
• How full the current conversation is, as a percentage of the context window.

Recommended: say yes. It's the easiest way to see at a glance that your clone's memory is alive. You can undo or change it anytime, and the Klyr theme also stays available on its own in the /theme picker.

Question 7: anonymized error reports

Klyr heals itself: when something breaks, it diagnoses and repairs what it safely can. This last question asks whether, when it hits a problem it can't fix, it may send an anonymized technical report. The error's signature only, never your notes, never your file paths, never a single memory. That way the fix ends up helping you and every other Klyr user.

Recommended: say yes. Anonymous means anonymous, and it makes Klyr better for you too. Change your mind anytime with /klyr:config.

The last step: bring it to life

What happens at the finish line depends on the memory choice you made in Question 1:

• Local: fully quit and reopen Claude Code one last time, so the memory tools load. Type /exit, then claude.
• Hosted: nothing to do. You're already signed in, and your clone is ready the moment setup says so.

Now make it prove it. Ask: "what do you know about me?" A generic assistant starts from zero. Your clone answers with your projects, your decisions, your way of working. That's the moment most people get it. And from here it compounds: every week it is a little more you.

Stuck somewhere?

Klyr is small and human: you reach the person who built it, not a ticket queue. Write to [email protected] and say where it stopped. Straight answers, no scripts.

Next: Your clone's memory.