Merge branch 'main' into integrations-opcua-logo-carousel

Author: Yndira-E

nuxt/pages/ai/index.vue

@@ -192,7 +192,11 @@ onUnmounted(() => {
          wash anchored at the top right, and an opaque hero sits exactly over it. The
          connector card below keeps its own white, because that is a card. -->
     <section class="w-full relative">
-      <div class="relative z-10 w-full px-6 pt-12 md:pt-24 pb-12 sm:pb-16">
+      <!-- pb-0: the recording below is the picker's own result, so the space between
+           them belongs to that section's pt rather than being split across a section
+           boundary here. Two paddings meeting in the middle of one thread is what
+           left 128px of empty gradient under the docs card. -->
+      <div class="relative z-10 w-full px-6 pt-12 md:pt-24 pb-0">
         <div class="sm:max-w-screen-lg mt-6 sm:mt-12 mx-auto">
           <div class="container m-auto text-left max-w-screen-lg">
             <div class="max-w-3xl mx-auto">
@@ -230,6 +234,64 @@ onUnmounted(() => {
       </div>
     </section>
 
+    <!-- DEMO: the payoff of the picker above. Those three steps say how to connect
+         an outside agent; this is what it looks like when you do, consent screen
+         included - the part a reader does not take on faith. It also hands off to
+         the governance section directly below, which is the same grant stated as a
+         claim rather than shown.
+         Autoplaying muted loop, the same treatment the install animation uses. The
+         controls stay on for one reason: at 57 seconds this runs long enough that
+         moving content needs a way to stop it, and the play/pause button is it.
+         Browsers pause an offscreen muted autoplay video by themselves, so there is
+         no observer here, and no preload hint either - autoplay settles that. The
+         poster still earns its place as the first paint before playback starts.
+         WebM only, and no MP4 fallback: full WebM support starts at Safari 16 and
+         iOS 17.4, which leaves 1.1% of global users out, and an H.264 copy costs
+         1.9MB to reach them - more than the WebM itself, on a video that now
+         autoplays for everyone. Those readers get the poster and a play button that
+         does nothing, which is the trade. Kept as a <source> with a type rather than
+         src on the video so a browser that cannot play WebM never fetches it.
+         The recording has no audio track at all, so muted is both the autoplay
+         precondition and the truth, and playsinline stops iOS taking it fullscreen.
+         pb-0 for the same reason the hero has it: these three sections are all
+         transparent, so a boundary between them is space and nothing else, and each
+         one owning only its top padding keeps that space to a single 64px. Doubling
+         up here put 128px above the governance card while the card sat 64px above
+         the capabilities band, which reads as the card floating low in a gap rather
+         than as a section change. Doubling is for boundaries that also change
+         background, the way the capabilities band does. -->
+    <section class="ff-ai-demo w-full px-6 pt-12 sm:pt-16 pb-0">
+      <div class="md:max-w-screen-lg m-auto">
+        <h2 class="max-md:text-center">Connect and build <span class="text-indigo-600">faster than ever</span></h2>
+        <p class="mt-3 text-gray-600 max-w-3xl">
+          Connect in seconds and start building. Use natural-language to leverage your company-sanctioned AI in a secure and governed way.
+        </p>
+        <!-- The frame is the wrapper, not the video, and it clips: Chrome builds its
+             native control bar in shadow DOM and ignores the video's own
+             border-radius, so a radius set directly on the video left the bar's
+             square corners poking out past the rounded border. Same rounded /
+             bordered / shadowed treatment the product page gives its hero shot.
+             A plain div rather than a figure now the caption is gone - a figure with
+             nothing to caption is just a wrapper. -->
+        <div class="mt-8 overflow-hidden rounded-lg border-2 border-indigo-100 shadow-2xl">
+          <video
+            class="block w-full"
+            poster="/images/ai/demo-external-agents-poster.jpg"
+            width="1600"
+            height="1056"
+            autoplay
+            loop
+            muted
+            playsinline
+            controls
+            aria-label="Screen recording: Claude adds FlowFuse as a custom MCP connector, signs in, is granted full access to one team, and then provisions a Node-RED instance from the OEE blueprint."
+          >
+            <source src="/images/ai/demo-external-agents.webm" type="video/webm">
+          </video>
+        </div>
+      </div>
+    </section>
+
     <!-- GOVERNANCE: the control plane every AI action runs through. Path-neutral,
          which is what lets it sit above the picker rather than under Expert. -->
     <div class="ff-ai-governance w-full px-6 py-12 sm:py-16">

nuxt/public/images/ai/demo-external-agents-poster.jpg

@@ -24,13 +24,19 @@ const GITHUB_HEADERS = {
 
 // Mirrors GitHub's own heading slugger so anchors written against a README's
 // GitHub-rendered preview (e.g. `#egm-optional` for "## EGM (optional)") keep resolving here.
+// GitHub strips punctuation first and only then turns each remaining space into its own
+// hyphen — it does not collapse runs of spaces. A heading like "Bugs / Feature request"
+// drops the slash but keeps both spaces around it, so it slugs to "bugs--feature-request"
+// (double hyphen), not "bugs-feature-request". Matching \s+ instead of a literal space
+// collapsed that run and produced anchors that never matched GitHub's own, breaking any
+// in-README link pointing at a heading with adjacent punctuation.
 function githubSlugify (heading: string): string {
     return encodeURIComponent(
         heading
             .trim()
             .toLowerCase()
-            .replace(/[^\w一-龥\- ]/g, '')
-            .replace(/\s+/g, '-')
+            .replace(/[^\p{L}\p{M}\p{N}\p{Pc}\- ]/gu, '')
+            .replace(/ /g, '-')
     )
 }
 

nuxt/public/images/ai/demo-external-agents.webm

@@ -0,0 +1,24 @@
+---
+title: FlowFuse Expert Upgraded to GPT 5.6 Luna
+description: FlowFuse Expert now runs on GPT 5.6 Luna, a frontier model that makes it markedly faster, smarter, and more capable at building your flows and taking action for you.
+date: 2026-08-31 12:00:00
+release: "3.1"
+authors: ["andrea-palmieri"]
+tags:
+  - changelog
+issues:
+---
+
+FlowFuse Expert now runs on GPT 5.6 Luna, one of the most capable AI models available today. It is a large step up: the Expert is noticeably faster, follows longer chains of work more reliably, and is simply smarter about turning what you ask for into the right result across your flows and your FlowFuse teams.
+
+This lifts everything the Expert already does. Building an application on the Node-RED canvas, taking platform actions on your behalf, and staying within the permissions you set: each of these gets sharper reasoning and quicker responses behind it. For you that means reaching the result you are after faster and with less back-and-forth, especially on larger, less-defined requests where the Expert has to plan, ask the right clarifying questions, and carry a task across several steps.
+
+It also builds directly on the work we have done [opening FlowFuse up to third-party AI agents](/docs/user/expert/third-party-agents/). Making the platform's automation and flow-building tools clear and reliable enough for any approved agent to use meant sharpening the tools themselves: better descriptions, cleaner inputs, and a growing, well-defined set of actions. The Expert calls those same tools, so every improvement we made for third-party tooling now lands in your first-class agent as well.
+
+The difference is that with your own agent the experience depends on the agent you bring, since each one has its own strengths. With the Expert we tune the whole experience end to end, model included, so this upgrade reaches everyone the moment it ships.
+
+We chose this model because it sits on the current frontier of what these systems can do: the rare combination of being highly capable, fast, and genuinely smart at the same time, rather than trading one off against the others. For an agent that acts on your platform in real time, that balance matters as much as raw capability.
+
+As the big pieces of AI enablement are now in place, we are moving into a more iterative rhythm: smaller, steady improvements that reach everyone using the Expert, released as we ship them. This model upgrade is the first of those.
+
+The upgrade is live now for all FlowFuse Expert users, on FlowFuse Cloud and Self Hosted, with nothing for you to change.