Website Agent

The website agent is Floe at the top of the funnel. Before a visitor is ready for a full demo, a compact Ask launcher carrying your site's name — Ask Floe on floe.so, which is how these docs refer to it — or a Talk to me presenter when you enable the video avatar, sits on your marketing site and answers their questions, grounded in your product. When they ask to see something live, the same agent hands off in place to a live demo. No new tab, no form.

It runs on the same session as the demo agent. A visitor can start by typing a question and finish by watching the agent drive your real product — one continuous conversation.

Answer questions before the demo

Not every visitor is ready to sit through a demo. Most have a question first: "Does this integrate with Salesforce?", "How is this different from Storylane?", "What does setup involve?"

The website agent answers those questions right on the page. A visitor clicks the launcher and the agent connects immediately, with the visitor's microphone off: it welcomes them, or answers directly when they arrive by typing a question or selecting a nudge. Answers come from your documentation, product content, and Ready Reference Guides — the same knowledge that powers the demo agent. It's the top-of-funnel conversation that used to require a rep or a form.

Grounded answers, honest about gaps

Answers are grounded in your ingested content and Ready Reference Guides. When the agent is confident, it answers. When it isn't — when a visitor asks about something you haven't documented or added directly — it says so and offers to connect them, rather than inventing an answer. A wrong answer on your marketing site is worse than an honest "I don't have that detail."

This is the same grounding discipline as the demo agent. Reference Guides can fill facts such as pricing or support, but they provide answers only; they never make the agent navigate or start a demo. Because answers are public, do not put secrets or customer-specific information in a guide.

Contextual nudges as they scroll

As a visitor moves through a page, the website agent can surface a short, contextual nudge tied to the section they're reading — a prompt like "Curious how scoring works? Ask me." Nudges are page-scoped: each one belongs to a specific page and section, so the prompt matches what's actually on screen and updates as the visitor scrolls. Clicking a nudge sends that question straight to the agent.

Nudge copy is generated from your content and prioritizes your buying-intent pages (pricing, product, comparisons). You control every nudge from the dashboard — edit the copy, or switch a page's nudge off entirely. See Website Agent Setup.

Answers shaped by who you sell to

The agent works from your ideal customer profile: a short summary and the buyer personas you sell to, with the problems each one feels. Floe drafts it from your content and you edit it. The agent uses it in two ways. It helps choose which question is worth asking a visitor: their role, their team, or the problem they have. And when a visitor's role or problem matches a persona, the answer leads with the part of your product that addresses that problem.

The profile never becomes a gate. During a conversation the agent does not score visitors, does not tell anyone whether they fit, and answers a visitor outside the profile as fully as one inside it. After the conversation ends, Floe uses the profile to give your team a lead score; the visitor never sees it. With an empty profile the agent behaves as it does without one, and leads are scored on engagement alone. Edit it under Website Agent → Configure.

Type or talk

The visitor's microphone starts off, and opening the agent does not request microphone permission. Opening with the main launcher or Talk to me starts a voice conversation: the agent welcomes the visitor aloud, and only once that welcome finishes does the browser ask for the microphone. On browsers that only allow the request during a tap, a one-time Ready to talk? button carries it instead. A visitor who declines — or whose microphone is blocked in browser settings or missing altogether — gets the text field straight away. Revoking the microphone mid-conversation stops the agent listening.

For a quiet text start, the launcher also carries a keyboard action, which opens the conversation with the agent's speaker muted and the same welcome in writing. Selecting a nudge sends that prompt as the first turn. Once the conversation is open, the visitor can type, mute or unmute the agent's replies, or turn their own microphone on. Either way it is the same agent and the same session.

Answers render as formatted text — lists, links, and emphasis — so a multi-part answer stays readable in the chat.

The avatar call surface

With a video avatar configured, the agent is a presenter standing on your page rather than a chat window that opens over it. While nobody is talking to it, the presenter waits a few steps back in the corner, with contextual nudges beside it. Choosing Talk to me brings it forward and the live session takes over in the same place; when the call ends it returns to waiting. Visitors who have asked their device for reduced motion see a still presenter instead.

Choose the avatar that should represent your agent. In languages with gendered self-reference, such as Hindi, the agent uses feminine forms for a female avatar and masculine forms for a male avatar, including during a live demo. Voice selection remains a separate setting.

The exchange appears in a bubble beside the presenter — what the visitor asked, spoken or typed, and the agent's reply. Under those words, Earlier expands the bubble into a scrollable transcript and, when a calendar is configured, Book a call opens booking beside the presenter without ending the call. A compact call bar carries the microphone, end-call, and controls to type instead or mute the agent's replies. On phones and tablets the presenter is drawn larger, with the bubble and the call bar stacked beneath it.

Sites without a video avatar are unchanged: the launcher opens the standard chat window with its composer and controls.

Hand off to a live demo, in place

This is what separates the website agent from a support chatbot. When a visitor asks to see something — "Can you show me?" — the agent doesn't dead-end at text. It transitions the same session into a live demo, opening your real product in a stage beside the chat. The conversation carries on while the visitor watches the agent drive, and they can expand the demo or minimize the chat to see more of it. With a video avatar in the bottom-right corner, on a desktop or a tablet held sideways, the conversation folds into a chat button in the call bar once the visitor has spoken to the agent, so your product gets more of the screen. Until then it stays beside the demo, so a visitor with the sound off can still follow along.

The stage opens the moment the hand-off begins, with a short card — your logo, when your site has one, and the line Opening the live product — while the agent brings up the first screen. The card stays up until the live picture is actually playing. From then on it gives way as soon as that first screen is up, or as soon as the agent starts speaking if that comes first; if the agent speaks before there is a picture, the card waits for the picture and gives way the moment it arrives. A visitor is never left looking at an empty stage. If your product takes a while to open, the card simply stays until it does; if it cannot be opened at all, the stage closes and the conversation carries on. The card's line is shown in the session's language.

There's no navigation away, no second form, no scheduling. The question becomes a demo in one continuous flow. From there it's the full demo agent: voice-led, driving your real UI, and guided by the same ideal customer profile.

The See it live action in the chat starts the hand-off and disappears once selected. With a video avatar the agent offers the demo in conversation instead. You can turn the demo hand-off off, or set the label on its call-to-action, from the dashboard.

Know your visitors

Visitors do not need to share an email to start. If you choose to, the agent can ask for an email (and optionally name, phone, company, and role) inline, at a moment you configure: after the first answer, after a chosen number of answered questions, or when the visitor shows demo intent. The ask never delays an answer, and when your embed requires consent it stays hidden until consent is granted.

When visitor data collection is enabled, Floe keeps activity from the same browser together so it can continue an active conversation across pages and tabs, and greet a returning visitor naturally. That browser identity is a continuity aid, not authentication: it does not follow a person to another device, and an email address alone never recalls another browser's conversation. See Security & Privacy and the SDK consent controls.

With visitor analytics on and the visitor's consent, Floe also records what a visitor does on your site before and after they talk to the agent: where they came from, the pages they viewed, how long they stayed and how much of each page they saw, the headings they stopped on, and the links, buttons and forms they used. It never reads a form field. Each lead's timeline in the dashboard is built from it; see Leads. What is read, what is left out, and how to exclude part of a page are in Security: What visitor analytics reads from the page.

By default the agent also uses that activity to choose what to offer, such as which part of your product to lead with and how much to explain. It is instructed never to tell a visitor what it saw. To turn this off, switch off Tailor the agent to what a visitor has read under Site → Settings → General → Visitor data collection; see Security: What the agent is given about a visitor.

Floe records which nudges are shown and clicked, and Website Agent → Nudges shows those counts for each nudge. See Website Agent Setup.

What reaches your sales team

A Website Agent conversation can produce the same internal follow-up as a full demo even when it ends before a live hand-off. A conversation qualifies when it produced an email address or concrete qualification or intent evidence — buying signals, budget, qualification answers, pain points, or product interest. A name, a page view, or generic Q&A by itself does not generate sales traffic.

Internal recaps label the lead Confirmed contact, Unverified email claim, or Anonymous lead, and by default only a confirmed contact is written to your CRM. See Integrations.

How it's different from a chatbot

A support chatbot answers text from a static knowledge base and stops there. The website agent:

Is grounded and honest — it answers from your product content and declines when it doesn't know, instead of guessing.
Is proactive — it surfaces contextual nudges tied to the page a visitor is on, rather than waiting to be opened.
Ends in a demo, not a transcript — when the visitor is ready, the same agent drives your real product live. The chat is the on-ramp to the demo, not a replacement for it.

Getting started

The website agent shares its knowledge with the demo agent, so setup builds on the same foundation:

Create a site for your product
Add content so Floe can answer accurately
Configure the website agent, which a new site has turned on — see Website Agent Setup
Embed Floe on your marketing site with the Browser SDK

FAQ

Can it answer questions before starting a demo? Yes — that's its main job. The Ask Floe launcher answers questions right on your marketing site, grounded in your product content, before a visitor commits to a full demo.

Does it show contextual nudges as I scroll? Yes. Nudges are page-scoped and section-aware: as a visitor reads, the agent can surface a short prompt tied to what's on screen. You control the copy and can turn any page's nudge off.

Can it hand off to a live demo in place? Yes. When a visitor asks to see the product, the same session transitions into a live demo — the agent opens your real product beside the chat. No new tab, no form.

Is the website agent the same as the demo agent? They're two phases of one agent on one session. The website agent handles the top-of-funnel Q&A; when the visitor is ready it becomes the demo agent and drives your product. Same knowledge, same conversation.

Do visitors have to talk? No. The visitor's mic starts off and no permission is requested on open. Opening with the main launcher or Talk to me makes the agent welcome them aloud; when that welcome finishes, the browser asks for microphone access, and declining puts the visitor straight into the text field. The keyboard action, a nudge, an opener, or the composer never asks for the microphone at all.

Does the chat remember a visitor across visits? A returning browser can continue an active conversation across pages and tabs. Once that conversation has ended, a later visit starts a new one and may pick up a relevant topic from that browser's earlier conversation. Floe never treats a browser identifier as proof of who is using it, and never uses an email address alone to recall a conversation from another browser or device.