Skip to content

No jargon, promise

Everything, explained properly.

Nobody is born knowing what a token is. Here is the whole thing in plain English, with pictures — and no assumption that you already know any of it.

Roughly a twelve-minute read. You can also just skip to the bit that is bothering you.

01 · What is AI?

What actually is AI?

A very well-read assistant that has never had an original thought in its life.

It has read an enormous amount of writing — books, websites, manuals, arguments on forums — and from all that it has got extremely good at one thing: working out what words should come next.

That sounds like a party trick. It is not, because to guess the next word really well you end up having to understand the question. Ask it to write a polite email declining a meeting, and it knows what polite looks like, what declining looks like, and what an email looks like.

What it is not: it is not alive, it does not want anything, and it is not secretly plotting. It is a tool that is very good with language, sitting still until you ask it something.

So can it be wrong?

Yes, and it will sound just as confident when it is. That is the one thing to remember. Treat it like a very fast, very well-read colleague who occasionally misremembers — useful constantly, worth checking on anything that matters.

IT HAS READbookswebsitesmanualsemailsargumentswhat comes next?a good answer

02 · AI worker

What makes it a worker and not just a chat?

It can actually do things, not just talk about them.

Most AI you have used can only produce words. You ask, it answers, and then you go and do the work yourself.

A worker is that same intelligence with hands. It can open your email and reply. It can put something in your calendar. It can go and read a webpage, save a file, or add a row to a spreadsheet. You are not getting advice about your inbox — you are getting your inbox handled.

And it does not stop when you close the laptop. It has its own machine that stays on, so a job you set at midnight is done by morning.

ORDINARY CHATreply to Priya?“Here is a draftyou could send…”…and now you do itA WORKERreply to Priya?“Sent. ✓”emailcalendarfilesit has hands

03 · Hermes & OpenClaw

What are Hermes and OpenClaw?

Two kinds of worker. Same idea — one of them built with more care.

When you create a worker you pick one of two. They are the same idea — a helper that lives on a server and goes and does things for you — but one of them is built with rather more care about what it is allowed to do.

OpenClaw came first. It is open source, it has an enormous community, and it drives a web browser by itself. Hand it something long and messy and it will go away and grind through it. That freedom is the whole appeal, and also the catch: a worker that acts on its own can get things wrong on its own.

Hermes is the same idea built for this job from the start. It comes from Nous Research and was designed from the start for a worker that lives on a server and acts for someone else, so the guardrails are part of the architecture rather than something added afterwards — layered permissions, approval checks before the things worth checking, and limits on what it can reach. It also remembers how it solved something and repeats it the same way next time.

That is why Hermes is our recommendation and the default. You lose very little, and you gain a worker that asks before it does something you would rather it asked about.

So should I ever pick OpenClaw?

If you want maximum independence and you are comfortable supervising it, yes — it is genuinely good at long, messy jobs. Give it only the connections it needs, check its work on anything that matters, and start it on something low-stakes. If that sounds like effort, use Hermes. Nobody will judge you.

Hermesrecommendedindependence: steadychecks in morequick to answerfew surprisesstart hereOpenClawmore independentindependence: highgoes off alonegreat at long jobswatch it at firstmore rope, more care

04 · Memory & skills

Does it remember me, or do I start again every time?

It remembers. That is the whole difference between this and a chat window.

You have probably used an AI that forgets you the moment you close the tab. You explain who you are, what you want, how you like it — and tomorrow you explain it all again. That gets old in about a week.

Yours keeps notes. As you talk to it, it writes down the things worth keeping: that you like short replies, that you never take calls before nine, that Priya is the client and not the supplier. Nothing dramatic — just the stuff you would otherwise have to say twice.

And it keeps a skills library. When it works out how to do something fiddly — chasing the people who have not paid, putting the Monday report together — it saves the method. Next time you ask, it does not work it out again. It just does it.

That library is yours. You can read it, change it, add ready-made skills someone else has written, or hand it a manual and say learn this. It all sits on your own server with everything else of yours.

The practical effect: month one you are explaining yourself, month three it already knows. The moment you notice is usually the day you catch yourself not repeating an instruction.

Can I see what it wrote about me?

Yes, and change it. It is a short, plain note on your server — not a secret profile in a database somewhere. If it has written something down that is wrong, cross it out.

every chatABOUT YOUkeep it shorthates 8am callsPriya = the bossSKILLSChase invoicesMonday reportTidy inboxnext timeit just does it

05 · Private server

What is a private server, and why do I get one?

Your own flat instead of a bed in a shared dorm. Same building, own front door.

A server is just a computer in a data centre that never gets switched off. Yours runs your worker, holds your files, and keeps your conversations.

The word people use for this is a VPS — a virtual private server. Ignore the jargon; the important word is private. Your worker is not one of thousands sharing one program and hoping the walls hold — it gets its own machine, with its own door. We hold the keys to the building, because we are the ones who build it and keep it patched. No other customer has a key to your flat.

That matters because of what you are handing it. Your email. Your calendar. Possibly your accounts. You want that living somewhere that is yours.

It is also why it is always on. It is not running on your laptop, so it does not stop when your laptop does.

Do I have to look after it?

No. You never see it, never log into it, never update it. We build it, run it, and keep it patched. You only ever see the chat.

SHAREDone room, one lock,everyone's thingsYOURSyour worker, your files,your door

06 · Model options

What are the model options?

Which AI brain your worker uses. It defaults to ChatGPT, and that is fine.

Your worker is not welded to one AI. It can run on any of a couple of dozen well-known ones — ChatGPT, Claude, Gemini, and plenty more — and you pick which, conversation by conversation.

Out of the box it uses ChatGPT, tuned for Hermes. You can leave it there and not be missing anything.

Why change it at all? Two reasons. Some models are far cheaper than others, so a lighter one makes your monthly credits stretch. And some are better at particular jobs — one is a stronger writer, another a stronger reasoner, another simply faster.

If you are technical: you can bring your own API key, or point your worker at a model running on your own hardware. That is covered on its own page.

There is also a thinking depth setting — Light, Balanced or Strong — which is the same idea in miniature. Light for sorting and deciding, Balanced for writing, Strong for the long careful jobs.

Two useful rules

Reading is cheap; writing is expensive — skimming a hundred emails costs very little, producing a long polished document costs a lot more. And the price of every model is shown right beside it when you choose, so you always know what you are about to spend.

See what each model is good at

Lightcheapestsort · spot · decideBalanceda few×write · plan · join upStrongdearestread long · reasonsame worker — you pick how hard it thinks

07 · Tokens & credits

What are tokens and credits?

Tokens are chunks of words. Credits are the money that pays for them. You will mostly ignore both.

AI does not read letters or words exactly — it reads tokens, which are word-sized chunks. "Calendar" might be one token. "Unbelievable" might be three. A rough rule: a hundred words is about a hundred and thirty tokens.

Everything your worker reads and writes is counted in tokens, and tokens cost money. Credits are simply your balance for that. Every plan comes with free credits each month — $2 on Basic, up to $30 on Max — and they reset when your month does.

Here is the part that matters: what a token costs depends on which model you picked. The same question can cost a fraction of a penny on the light setting, or tens of times that on the heaviest one. You are choosing the price every time you choose the model — and when you open the model list, the price of each one is sitting right there next to it. Nothing is hidden.

For a lot of people none of this ever comes up. You leave it on the default and you rarely think about it again.

If you do use more, you top up: $10, $25, $50 or $100, and bought credits never expire. Auto top-up is on by default so your worker never goes quiet mid-job — $10 whenever it drops below $1, up to $50 a month, and you can turn it off.

Why the 5%?

Top-ups carry a 5% processing margin — pay $10 and $9.52 lands as credit. That is the card fee and the handling, said out loud rather than hidden in a footer.

FREE, EVERY MONTH$2 – $30resets on your billing day(does not roll over)TOPPED UP$10 – $100never expiresonly used once thefree ones run out

08 · Where you message it

How do I actually talk to it?

In your dashboard, or wherever you already text, like it is a person.

The moment your worker is running there is a Chat button next to it. Press it and you are talking. That is the whole setup.

But nobody needs another tab. So you can connect it once and your worker turns up where you already message — Telegram, a plain text message, Slack, and more — as a normal conversation. You message it the way you message anyone. It replies in the same thread.

It is the same worker either way, with the same memory. Ask it something from your phone while you are out, open the dashboard when you are back at your desk, and it knows exactly what you were talking about.

Whose account is it?

Yours. You connect your own accounts, and you can disconnect any of them in one click. Lionox is an independent service and is not affiliated with WhatsApp, Meta, Google, or any messaging platform.

Ivyone workerone memory, one private serverWhatsAppTelegramSlackDashboardsame conversation, wherever you open it

09 · What is a CRM?

What is a CRM, and why would my worker touch one?

A shared address book that remembers every conversation. Your worker can keep it fed.

CRM stands for customer relationship management, which is a terrible name for something quite simple: a list of the people you deal with, and a record of what happened with each of them.

Who they are, when you last spoke, what they wanted, what you promised, what stage it is at. If you have ever lost a deal because nobody followed up, a CRM is the thing that was missing.

You may already have one — Lionox CRM, Salesforce, HubSpot, Pipedrive, Attio and plenty more all do this job. Your worker connects to them the same way it connects to anything else.

The problem with every CRM ever built is that somebody has to type into it, and nobody does. That is where a worker earns its money. It reads the email thread, updates the record, logs the call, sets the follow-up — without being asked twice.

If you do not use one, this is probably not your section, and that is fine.

Priya Ramanlast spokeTuesdaywantedQ3 quotepromisedrevised pricestagedecidingnextfollow up Frikept fed bythe email threadyou already hadyour workertypes it up, not youthe bit nobody ever does

Still not sure? Ask it.

Genuinely — the fastest way to understand what it does is to give it something small and watch. Three minutes, and there is a 30-day refund if it turns out not to be for you.

Or read how it works and what it costs.

Ready to meet your AI worker?

Sign up, press one button, and start chatting. You will be talking to it in about three minutes.

Starting at only $19.99/mo · Ready in minutes · Cancel anytime

30-day money-back guaranteePlan only — not credits already used
  • Ready in minutes
  • No tech skills needed
  • Cancel anytime