support for your site

How it works

AISupportForge puts a custom AI support agent on your website. It answers your prospective customers before they buy, helps your logged-in customers with answers about their own account, and can see what a confused visitor actually did, so nobody has to type "can you describe what you were doing?"

[01] // forging your agent

Built from your knowledge, not a script

No decision trees to draw, no canned replies to write. You talk; the forge listens.

STEP 1

The AI interviews you

A guided conversation learns your business: what you sell, your policies, the questions your customers actually ask.

STEP 2

It reads everything

Upload docs, FAQs, and policies, or point it at your site and it reads your pages, including the buttons, forms, and flows your visitors use.

STEP 3

Test-drive before launch

Try your agent in a private live preview. Ask it hard questions, refine what it knows, and only ship it when it earns your trust.

STEP 4

One line of code

Paste a single script tag into your site. The chat bubble appears, and your agent is on duty 24/7.

<script src="https://aisupportforge.com/w/your-site-id.js" async></script>

[02] // one agent, two jobs

Prospects and customers get different help

The agent knows the difference between a stranger deciding whether to buy and a paying customer who needs a real answer about their own account.

visitors & prospects

Pre-sales mode

For anonymous visitors, the agent is your best salesperson, instantly answering from your public knowledge:

Pricing, features, and "how does this compare?"
Objections handled at the moment of doubt, not after they leave
When it senses buying intent or hits a question it can't answer, it offers to take their email, and the lead lands in your inbox

logged-in customers

Support mode

When your site tells the agent who's chatting, via a cryptographically signed token only your server can create, support gets personal:

"Your subscription renews on March 4th."
"Order #1123 shipped yesterday, here's the tracking."
Account answers come from your own data, read-only and scoped strictly to that one customer

[03] // smart diagnostics

It sees what any visitor was doing

No more "what were you doing when it broke?"

As a visitor browses, the widget keeps a short trail of their recent steps (pages viewed, buttons clicked, form errors hit, even frustrated rapid clicking) held privately in their own browser. The moment they open the chat and ask for help, that trail arrives with their question:

14:02:11 viewed /billing 14:02:29 clicked "Update Card" 14:02:31 form error: "ZIP code required" 14:02:38 clicked "Update Card" ×4 (frustration) 14:02:44 opened chat: "why won't my card save??"

So instead of twenty questions, the agent answers on the first try: "Your card didn't save because the ZIP code field was left empty; fill that in and Update Card will go through."

Nothing to switch on, and it is not just for people who are logged in. This works from the moment the agent is on your site, for every visitor, including a stranger who has never bought anything and never will. There is no setting, no extra code and no extra cost.

It is worth not confusing this with recognising your logged-in customers, which is a separate and genuinely optional feature. This one tells the agent what someone just did. That one tells it who they are, so it can answer about their own account. They stack, and neither needs the other.

[04] // silent until spoken to

Privacy is the architecture, not a checkbox

The widget's operating rule: nothing about a visitor leaves their browser until they choose to start a conversation. That's not a policy promise, it's how it's built.

No constant connection

The widget doesn't stream behavior to our servers. Activity stays on the visitor's device and is only sent, by them, when they ask for help.

No cookies, no fingerprints

Anonymous visitors get no persistent identifier. A throwaway session ID exists only while a chat is open and dies with the tab.

Never your customers' secrets

What visitors type into form fields is never recorded, only that a field errored. Passwords, card numbers, and personal data stay where they belong.

[05] // humans stay in the loop

When the AI shouldn't answer, it doesn't

Graceful hand-off

If the agent isn't confident, or someone is getting frustrated, it stops guessing. It gathers the issue, emails your support address with the full transcript and activity trail attached, and tells them a human will follow up. Nobody hits a dead end, whether they are a customer or somebody still deciding.

Your command center

Every conversation, captured lead, and escalation lands in your dashboard. The most valuable report: "questions your agent couldn't answer", a running list of exactly what documentation to add next. A weekly digest keeps score: "143 conversations, 12 leads, top unanswered question: refunds."

Ready when you are.

Start free with 50 messages a month. No card, no sales call, and you can see your agent answering questions about your own site within a few minutes. Create your account or look at the plans.