No AI extension writes this page. It is nonetheless every buyer's first question: how much is this going to cost me?
Short answer: your Astus subscription, and nothing else — unless you use the chat.
Two meters, not to be confused
The Astus subscription gives you the product and its updates. It is paid once a year and does not depend on your usage.
Your AI provider's credit — Google, Mistral or Anthropic — is paid to them, not to us. Astus adds no margin, resells no token and has no visibility on your bill.
You can perfectly well use Astus with no AI account at all. You lose the chat; you keep everything else.
What never consumes a token
Everything shown on screen is computed by Astus itself, on your own server or by free Google services:
| Feature | Cost |
|---|---|
| The four site audit scores | free (Google PageSpeed) |
| The Joomla configuration audit, and its fixes | free |
| SEO analysis of an article, a category, a menu item | free |
| Google Search Console figures | free |
| Broken link checking | free |
| The list of images with no alternative text | free |
| The full report and its printout | free |
These work even when no AI key is set. You can measure your site every day for a year without spending a penny with an AI provider.
What does consume
The chat, and only the chat. Every time you press Send, a request goes to your provider and they bill you for it.
The Fix and Ask Astus buttons scattered through the reports consume nothing by themselves: they write the request into the input box and stop there. Until you press Send, nothing has left.
That is a design decision, not an ergonomic oversight. A button that triggers spending should be a button you press knowingly.
What a token is
A token is a piece of a word. A hundred English words weigh roughly 130 tokens. A request has two meters:
- tokens sent — your message, the recent conversation history, the SEO charter and the description of the tools Astus has;
- tokens generated — the answer.
The second cost markedly more than the first, with every provider.
Beneath each reply, Astus shows exactly what was just consumed: number of calls, tokens sent, tokens generated. Nothing is hidden, and you do not need to open your provider's console to find out.
Caching, which divides the bill
The start of every request is always the same: the SEO charter and the tool descriptions. Providers know this and bill that part up to ten times less when they recognise it.
Astus shows that share beneath the reply: "of which N cached". The higher it is, the cheaper the request. It is also why chaining several requests in the same conversation costs less than opening a fresh conversation each time.
What pushes the bill up, and how to avoid it
Steps. One request can chain several tools — read the article, analyse it, fix it, analyse again. Each step is a billed call. The Max steps per request setting bounds that chain; lowering it protects you from a request that loops, raising it allows complete cycles.
Talkative models. A model that "thinks" at length generates a lot of tokens. If your bill surprises you, change model in the settings before changing your habits.
History. Astus sends the last twelve exchanges. A conversation that drags on costs more with every turn: the New conversation button is not decoration.
Choosing a provider
All three are configured in exactly the same way, and swapped in two clicks.
Google Gemini — free tier with tight quotas, or paid tier if billing is enabled on the project. It is the same key either way, only the quotas change. Note: the quota is counted per Google project, not per key.
Mistral — free tier, one request per second. A French AI, if data hosted in Europe matters to you.
Anthropic Claude — pay as you go, with no blocking rate limit. A Claude Pro subscription does not grant API access: you need a key from console.anthropic.com.
And the PageSpeed key?
Free, no credit card, 25,000 measurements a day. It has nothing to do with AI — it is a key for a different Google service, and a Google AI Studio key will be refused here.
Without it Astus still measures, but Google heavily limits how many measurements you get per hour. See Google keys, and declaring your property.