The assistant says: “Hi — ask me anything about this page.”
AI voice onboarding for SaaS
OnboardingVoice is an AI assistant that lives inside your web app — it answers out loud, and points at the exact button, in real time. Add a lightweight widget. Connect your docs. Help users find their next step.
Chatbots tell your users where to click.
OnboardingVoice points.
Support articles describe. Tooltips interrupt. Your users are already in the product — they don’t need directions read back to them. They need someone to walk over and show them.
A product walkthrough
An illustrated preview of the experience. The assistant speaks its answer while a spotlight finds the relevant element — then walks the user through, step by step.
Outstanding
$12,450
Paid this month
$8,920
Drafts
4
Recent invoices
| Invoice | Client | Amount | Status |
|---|---|---|---|
| #0043 | Nova Labs | $1,150.00 | Paid |
| #0042 | Brightpath | $960.00 | Sent |
| #0041 | Corley & Co | $2,300.00 | Overdue |
| #0040 | Fielder Studio | $780.00 | Paid |
New invoice
DraftClient
Amount
Due
User asks: “How do I send my first invoice?”
Happy to. See the glowing button, top right? That’s New Invoice — I’ll walk you through it.
Source: Ledgerline Docs ↗
Every highlight, click and keystroke above is the widget’s actual guidance engine — eight visual action types, chained into tours.
Four sample products. Ask a question, then follow the guidance on screen.
From zero to guided
01
Point the no-code Studio at your product. It learns your screens automatically — URLs, page structure, even single-page apps — and lets you mark the elements that matter.
02
Upload your docs, help center, or guides. Answers come grounded in your own content — with citations — never generic model guesses.
“Where do I change the billing owner?”
Under Settings → Billing → Owner. Only admins can change this.
Source: Help center ↗03
Paste a single tag — three config fields, ~58KB gzipped for the core. The mic appears in your product. No rebuild, no code changes.
<script src="https://cdn.onboardingvoice.com/widget.js"
data-tenant="your-app"
data-key="pk_live_…"
data-position="bottom-right">
</script>Under the hood
Real products ship weekly, run as single-page apps, and hide half their UI in shadow DOM. The guidance engine is built for exactly that.
Every element is stored with fallback selectors and a structural signature. If a selector fails, the widget re-scores candidates — piercing shadow DOM when needed — and still finds the right target.
Real-time speech-to-speech over WebRTC for natural conversation. Browser speech for instant, on-device response. Server speech for consistency at scale.
Detects where the user is — URL patterns and DOM sentinels, single-page-app aware — so guidance always fits the current view.
history.pushState → screen:changed
Chained into full guided sequences — tours that click, type, and scroll for the user.
Retrieval over your own docs, with citations users can open. If it isn’t in your content, it says so — no invented answers.
Source: your docs ↗~58KB core, gzipped
3 required config fields
Database-per-tenant isolation
Built-in guidance analytics
Design partner program
We instrument your app ourselves — you pay one flat pilot fee and get a rigorous, holdout-controlled measurement of what changes. Your data stays isolated the whole time: an architecture guarantee, not a policy promise.
Three tenants. Three databases. Zero lines between them — by architecture, not by policy.
Early access
We work hands-on with a small group of products before general availability. Founding partners get concierge setup and pricing that never expires.
Concierge onboarding
we map your app with you, screen by screen
A direct line to the founding team
your feedback shapes the roadmap
Founding pricing, locked for life
it never expires, whatever we charge later
No credit card. We reply to every request personally.
Asked, answered
One script tag with three config fields. There's nothing to rebuild and no changes to your app's code.
Every element is stored with fallback selectors and a structural signature. When a selector stops matching, the widget re-scores candidates — piercing shadow DOM when needed — and still finds the right target. If nothing scores high enough, it degrades to a spoken answer instead of highlighting the wrong thing.
From your documentation only. You upload docs, help-center articles, or guides in the Studio; answers are retrieved from that content and returned with citations users can open. When your content doesn't cover a question, the assistant says so instead of guessing.
Yes. The widget is a framework-agnostic web component. Screen detection is SPA-aware — it watches route changes, URL patterns, and DOM sentinels — with router adapters for React Router, Vue Router, and Next.js, plus shadow-DOM piercing for component libraries.
Each customer runs on a physically separate database — your content, transcripts, and analytics never share a table with anyone else's. API keys are stored hashed, and the widget ships no secrets to the browser.
No. Voice is optional — the widget also takes typed questions in the same panel, and the visual guidance works identically either way. Three speech modes (real-time WebRTC, browser speech, server speech) are there when voice fits the moment.
The onboarding journal
Practical ideas for product teams: shorter paths to value, more useful conversations, and onboarding that meets users where they are.
Article
13 February 2026 · 9 min read
Guide
12 February 2026 · 10 min read
Guide
10 February 2026 · 9 min read
Contact
Tell us about your product and where new users get stuck. We’ll show you what a voice that points can do about it.
Prefer email? [email protected]
The assistant says: “Still reading? Let me show you instead.”
Join the waitlist — early partners onboard first.
