Questions organizations ask us

No cloud AI at all. The Guide never generates text — it chooses from a script you approved and an index of your real pages, using deterministic matching that runs in the visitor's browser. The full mechanism, including the deeper technical detail, is written up openly in How it works. The optional "understanding" upgrade adds a small, named, pinned open-source model that also runs on the visitor's device — it re-ranks your pages by meaning and cannot add a single word to what the Guide can say.

Every sentence, yes — written during onboarding and approved by you (plus small fill-ins like the visitor's own name, when they've offered it). What's computed is the choice: which line, which of your pages. That split — authored voice, computed understanding — is the whole design, and it's why the Guide can go where generative chatbots legally can't.

It says so, warmly, and offers directions — the categories of your site, or a hand-off you choose (your contact page, your box office). It never bluffs. Misses are also collected into a monthly "asked and didn't find" report: a to-do list for your content, from your visitors' own words.

We read your public pages once, politely (identified crawler, slow, respecting your robots rules) and keep only facts: titles, addresses, headings, description lines. Nothing is copied or rehosted — every card opens your real page. When your site changes, we re-read on a schedule.

It can only say lines from your approved script and offer pages from your index. The worst case is a wrong page suggestion or an honest "I don't have that." There's no generator to jailbreak and no answer to retract. Your team can read the complete script — "what can it say?" has a finite, printable answer.

Safety runs before everything else. Messages that signal distress — crisis, bullying, abuse — are recognized by a dedicated layer that drops the tour-guide act and offers real, verified help lines, in the visitor's language. That layer is tested by an automated regression suite in both English and French before every release, and it never depends on an AI model. This is why family brands can ship the Guide when they can't ship a chatbot.

You're expected to — the script is yours. We draft, you edit and approve, nothing ships without sign-off, and after launch the miss report tells you what to add next. Many teams treat the script like site copy: owned by brand, reviewed like any page.

The neutral character in the demos is a placeholder. Yours gets your colours, shape, name and voice — and you own the character design (the animation engine stays ours, licensed to you, including for use in your marketing). Your existing mascot can also be brought onto the rig.

Never to us: there is no chat server, and SpecEnvoy never sees transcripts or visitor data. If you enable measurement, usage events (asked/routed/missed) go into your own analytics under your privacy policy — and the guide's own privacy answer to visitors says so. Visitor memory (a name, a preference) stays in that visitor's browser; one phrase wipes it.

The Guide processes no personal information server-side, sets no cookies, and calls no third-party services — its privacy exposure is essentially that of a static web page. Visitor-side memory uses browser storage for functionality only. We'll gladly walk your privacy officer through it; the honest review is short.

It can run entirely on your own domain: the Guide is static files with no external calls, so your Content-Security-Policy can stay strict and there's no vendor endpoint to allow-list.

Real. French has its own script (authored in French, vous-form by default), its own question understanding, and — most importantly — its own safety layer, built and tested in French with the right francophone help lines. Same engine, first-class French brain. Try both: English demo · démo française.

English and French today; the engine also runs a Chinese brain on our own site. Each new language gets the same treatment — authored script plus a language-specific safety layer with its own test suite — so we add them deliberately, not as translate widgets.

One <script> tag on your existing site — any CMS, any stack, no backend work. You can trial it on a staging page first, or even preview it on your live site in a browser console during a call.

The bundle is small (~150 KB, loads deferred) and the character waits quietly in the corner — no popups, no autoplay, dismissible, and it stays closed if the visitor closed it. It also keeps the conversation as visitors move between pages.

The same as when it doesn't: nothing per message. The intelligence runs in each visitor's browser, so a viral day costs you zero extra — there is no usage meter anywhere in the product.

The widget targets WCAG 2.1 AA (keyboard operation, screen-reader labels, reduced-motion respect); a formal audit pass is part of pilot delivery, and we only claim compliance once it's verified — we won't hand you an accessibility statement we haven't earned.

Measurement runs in your analytics, never ours. The guide emits named in-page events (opened, asked, routed, missed, card clicked); with your opt-in they flow into the analytics you already run, under your privacy policy — SpecEnvoy receives nothing, and visitors gain no new tracker. The agreed metric is computed from those events; the monthly report is our written reading of them: the metric against its target, top asks, and the misses turned into a content to-do list. Details on the pilot page.

90 days · one success metric agreed in writing · flat CA$1,500 including a custom character skin · biweekly 30-minute check-ins · cancel anytime · you keep the data and the findings. After a successful pilot, ongoing service (hosting the pack, scheduled re-crawls, script updates) is a modest monthly fee — never usage-based. The full program, timeline and fair-exit promise: the pilot page.

Your Guide keeps working, untouched — it's static files already on your infrastructure, with no server of ours in the loop. Annual agreements include a source-escrow clause: if we cease operations or discontinue support, you receive the source with a perpetual license to maintain your own deployment.

You own your script, your pack, and your character's design. We own the engine and license it to you. Your data was never ours to begin with — there isn't any on our side.

We're an Ontario supplier sized for direct/discretionary purchase at pilot scale, and we'll complete your supplier forms. Insurance certificates can be arranged for contracts that require them — ask early and we'll sort it before paperwork becomes the bottleneck.

Honestly: you could build a chatbot quickly — that's the thing your legal and brand review likely already declined. What's slow to rebuild is what makes this shippable on your brand: a two-language safety layer refined across hundreds of tested edge cases, an animated character with no art pipeline, deterministic behavior your counsel can sign off on, and the ongoing curation. We've published how it all works precisely because the value isn't a secret trick — it's the finished, tested, maintained thing, at a pilot price far below the specialist time to recreate it.