Widget Settings
The widget is deliberately small: no chat library, no framework, nothing loaded from a third-party CDN. It is a floating button that opens a panel, streams the reply as the model writes it, and gets out of the way.
Appearance
Colour
Pick a primary colour, or click one of the swatches taken from your theme's own palette. Rather than trusting that the result will be readable, the plugin measures the colour's relative luminance and derives the header text, bubble and accent colours from it so they clear WCAG AA contrast. A dark brand colour gets light text; a bright one gets dark text; you do not have to think about it.
The settings screen previews both the closed button and the open panel using the real widget markup, so what you see is what visitors get.
Position
Bottom-right or bottom-left. If you already have something in a corner (a cookie notice, a back-to-top button, another floating widget), put the chat button in the other one and check on mobile, where the collision is usually worse.
Content
Bot name
Shown in the panel header. Something that reads as a helper rather than a person tends to work better; a human first name invites visitors to assume they are talking to staff.
Greeting
The message shown when the panel opens. Keep it to one line, and make it about what the bot can do rather than a generic hello. "Ask me about shipping, returns or sizing" gets far better first questions than "Hi! How can I help?", because most visitors do not know what a bot on a site they have never used is able to answer.
Suggested questions
Up to six, one per line. They appear as buttons in the panel and are the highest-leverage setting here. Use the questions your support inbox actually receives, and make sure your content genuinely answers each one. A suggested question the bot then fumbles is worse than not suggesting it.
Human support link
Where the bot points people when it cannot help. Set this. The decline path is the difference between a bot that is honest and a bot that is a dead end, and without a destination the honesty has nowhere to go.
What the widget renders
Replies stream in token by token. The widget escapes everything first and then re-applies a narrow subset of Markdown: bold, inline code, fenced code blocks, headings rendered as bold, line breaks, and links restricted to http, https and root-relative URLs. This is a security boundary, not a styling limitation: the model's output is treated as untrusted text.
Known rough edges in beta
- Markdown lists and tables are not converted, so they can appear with literal asterisks or pipes. If the bot keeps producing lists, mention the preferred format in your site facts.
- Very long answers can stop mid-sentence, because no explicit length limit is passed to the model.
Browser support
The widget uses streaming fetch, which means current versions of Chrome, Edge, Firefox and Safari. There is no fallback for browsers without it; on those, the widget will not work rather than degrading. In practice this affects a very small share of traffic, but if you serve an unusual audience it is worth knowing.
Where the widget appears
Site-wide once enabled, and there is no per-page control in the settings screen yet; that is on the roadmap. Developers can filter its visibility in code with flarepipe_chat_widget_visible.
The badge
A small "Powered by Flarepipe" link sits under the panel. It can be removed with the flarepipe_chat_show_badge filter.
Turning it off
The enable checkbox in settings removes the widget from the front end without touching your index or your Cloudflare resources. Use it while you are still tuning content, so visitors do not meet a half-trained bot.