Add-on service
It sits in the bottom corner, guides visitors to the right destination and asks three questions instead of twenty. Every answer it gives was written in advance — which is why it needs no consent, costs nothing per conversation and never goes down.
People who land on a services page rarely have a question — they have a situation. Too many options, no price they can place, and no appetite for filling in a contact form before they know they are in the right place. So they leave. A companion inverts that by taking the role a person at the door takes in a shop: it asks what is being looked for and then shows exactly one way forward. Not six. And if someone does not want to talk, it stops talking — that is a rule in the code, not just an intention.
In use
This platform: 297 prerendered pages, two languages, one companion
The companion you can see here is the same product I deliver — not a demo build. It runs on 275 pages of this platform, knows the difference between a glossary entry and a services page, and names the subject of the page it is on. Open it, turn it down once, and watch what happens next: nothing. That is the part you have to try, because it cannot be claimed.
The example shown is my own product on my own platform — an in-house build, not an independent client project. I say so rather than leave it out. The companion is new; there is no client installation to point at yet, and I do not invent one.
Why it is built differently
Every answer exists as text in the project and is signed off by you. A language model will produce something plausible when in doubt — in front of a law firm or an engineering office that is not a detail, it is a liability question.
Nothing leaves the device. The session store carries no identifier, only counters, and is discarded when the tab closes. So it touches neither the consent banner nor a data-processing agreement.
Zero bytes in the initial bundle, measured and protected by a build gate. It loads after the first render — the page is complete before it arrives, not after.
At most three utterances per visit, at least two minutes apart, and a refusal both lowers the allowance and locks the topic. Those limits live in the code and pass through a single gatekeeper — no trigger gets its own way around them.
Process
We collect every destination a visitor could be guided to — and cut until at most six remain. That session is the real value: anyone who cannot name six destinations does not have a companion problem, they have an offer problem.
From stage 2 on, the decision tree takes shape: at most three questions before a recommendation, and every recommendation gets a sentence that justifies it. Anyone who has to answer four times would have been faster on the phone.
The companion takes on your site’s colours, spacing and radii through its own design tokens. It does not look like a bought-in widget, because it is not one.
Every path is clicked through, keyboard operation and screen reader are checked, three screen sizes measured. Afterwards you get the file that holds the greeting, the destinations and the questions — you can change it yourself.
Stages
The stages build on one another and share the same state machine. Upgrading later therefore costs the difference, not the full price — nothing is rebuilt, something is added.
It is there, and it guides. For sites where visitors cannot find the one right destination.
on request about one week
It asks instead of merely showing. For offers that need explaining before anyone buys.
on request two weeks
It knows where it is standing. For sites with many subpages, where one sentence for all of them will not do.
on request three weeks
The capsule becomes a person. Only worthwhile where presence itself is the product.
on request five to six weeks
Open questions instead of fixed paths. Possible, but bound to three conditions I will not skip.
on request once the conditions are cleared
Setup, additional languages, and later expansion are quoted according to the agreed scope. Ongoing care is available but optional: stages 1 to 4 create no operating costs — no server and no per-conversation fee — so there is no mandatory subscription. Optional care is currently priced on request only.
Questions
In stages 1 to 4, no. It is a decision tree: it asks up to three questions and then recommends one of your destinations. That difference is not a cut-down version, it is the reason it runs without consent, without running costs and without liability risk. A language model is provided for as stage 5 and serves the same interface — so this is not a dead end, it is the order of things.
Not for stages 1 to 4. Nothing leaves the device, no identifier is issued, and the session store holds only counters that disappear when the tab closes. Storing the language choice permanently asks first, and if you withdraw consent the entry is deleted and no longer read. Stage 5 would be a different matter — which is why it sits behind three conditions.
No, and that is measured rather than claimed: the companion is in none of the bundles loaded on first request — zero bytes. It is fetched only after the first render. A build gate breaks the build if that ever changes.
A plain link to your contact route sits in exactly the same place. That is not a stopgap but the delivered state of every prerendered page — the capsule replaces it only once the browser is ready.
It never opens by itself. It says something at most three times per visit, at least two minutes apart, and a refusal lowers that allowance further and locks the topic for the rest of the session. All triggers pass through the same checkpoint, so no future trigger gets its own way around those limits.
Yes. The greeting, the destinations and the questions live in one readable file that you receive at handover. If you would rather not touch it, add the care option — both are possible, neither is required.
Then it costs the difference between the stages. All stages share the same state machine and the same interfaces — an upgrade adds something, nothing is thrown away and rebuilt. That is precisely why starting at stage 1 is not a wasted investment.
On an Angular site, yes — there it is a configuration file and one small adapter file. For other systems, WordPress for instance, I look at it first and tell you before quoting whether it works. I do not sell an integration I have not checked.
That is the only question that has to be answered before a companion. We go through it in one conversation — free of charge, and if it turns out a visible contact button is enough, I will say so.
Request a first conversation