Not a Chat-bot, a Do-bot
Ablehand is a command bar for your web app. A user types what they want, like "invite a teammate" or "export last quarter," and it gets done in their own session, in seconds, inside guardrails you set. One script tag. No tours to author, no selectors to maintain.
What changes for your team.
"How do I" answered in the product
The questions that fill your support queue become goals the user types once and watches complete. Settings, billing, invites, exports, cancellations.
Users reach the feature by asking
New users don't hunt for the feature; they ask for it. Onboarding steps happen in the product, in their session, without a tour you have to maintain.
You learn what users wanted and didn't get
Every goal is counted. The Needs page and a Monday digest show what people asked for, what resolved, where they got stuck, and the first fix to try.
Built to be measured.
Measured on production, not a slide.
~0.5 s
per step
Median 526 ms round trip from the browser on production over the last 30 days, of which the decision itself takes 268 ms. Users see the next step before they'd have finished reading a tooltip.
0
credentials leave the page
No cookies, passwords, or page screenshots. Card numbers, SSNs, keys, and tokens in page content are masked in the browser before it's sent.
Any
stack, any language
It reads roles and labels on the live page, not your code. React, Vue, Rails, PHP, server-rendered, and whatever language your users speak.
No authored tours. Ablehand reasons over the live page.
A fast structured-decision model reads what's actually on screen and chooses one observed element per step, so it adapts when your UI changes and covers tasks nobody scripted.
Snapshot
The widget indexes the visible, interactive controls on the current page: labels and roles, never selectors or code.
Decide
The edge asks the model for one choice: which observed control, and what to do. Click, type, select, or navigate.
Act
The widget re-validates that choice against the live DOM and performs it in the user's session, then repeats until the goal is met.
Guide, or Do. You set the line.
Default to showing users the way; let them opt into hands-free. Irreversible steps ask first.
Guide
Highlights each step and narrates it; the user performs it. Nothing happens that the user doesn't do.
Do
Completes the task for the user ("cancel my plan", "export last quarter's invoices"), confirming consequential actions.
Hybrid
Does the reversible steps, guides or confirms the ones that matter. Your capability gates bound what's possible.
Safe by construction.
Same-origin, in-session, loaded by a tag you control. The security model is the architecture, not a policy.
It can only do what your user can
Runs in their own browser session through the page's own controls, with no credentials of its own. What leaves the page is listed, field by field, on the trust page.
The model never emits a selector or URL
It returns an index into controls the widget already observed and re-validates, so a wrong answer can't point it somewhere arbitrary.
Secrets stay in the browser
Password, payment, and hidden fields are never read; cards, SSNs, tokens, and keys in page content are masked before any snapshot is sent.
You hold the guardrails
Per-site capability gates, deny lists, confirmation policy, and operator hints, edited in the console without shipping code.
Full detail in Security & trust model and Trust & data handling, including the subprocessor list and retention policy.
One line to deploy.
Add the tag and register your origins; once your account is approved, the same widget running on this page runs on your app.
<script src="https://ablehand.ai/nav.0.x.js"
data-site-key="pk_your_key"
data-mode="guide" defer></script>
Why not a product tour or a server agent?
| Scripted product tours | Hosted browser agents | Ablehand | |
|---|---|---|---|
| Authoring | Hand-built flows per task | Prompts + a hosted browser | None. Reads the live page |
| Survives a UI change | Can break when the UI changes | Varies | Adapts |
| Where it runs | Overlay on your page | A remote browser | The user's own session |
| Credentials | N/A (only points) | Needs the user's, or a hosted login | Never leave the browser |
| Completes the task | No, only shows | Yes, off-site | Yes, in-session |
See it drive a real app.
The demo is a working SaaS workspace. Open the bar at the bottom of the page and try "turn off the email digest", "open my billing settings", or "search products for Echo". Or use the bar on this page to go there.
Open the live demo →