wiki-plugin-similarity 0.1.0

package/client/similarity.js

@@ -0,0 +1,38 @@
+/* wiki-plugin-similarity - 0.1.0 - Tue, 09 Jun 2026 19:28:16 GMT */
+(()=>{var j={high:.78,medium:.68,low:.58},v=j.medium,E=10,R=t=>{let e=[],s=null,n=null,i="search";for(let c of t.split(`
+`)){let o=c.trim();if(!o||o.startsWith("#"))continue;let a=o.toUpperCase();if(a.startsWith("SIMILAR:")){let p=o.split(":")[1].trim().toLowerCase();s=j[p]??v,!e.length&&i==="search"&&(i="similar");continue}if(a.startsWith("THRESHOLD:")){s=parseFloat(o.split(":")[1])||v;continue}if(a.startsWith("LIMIT:")){n=parseInt(o.split(":")[1])||E;continue}e.push(o)}return{mode:i,specs:e,threshold:s??v,limit:n??E}},z=t=>t.includes("*")||t.includes("?"),F=t=>t.toLowerCase().replace(/\s+/g,"-").replace(/[^a-z0-9-]/g,""),L=new Map,U=async(t,e)=>{if(L.has(t))return L.get(t);let s=`${e}/system/indexed-domains.json?pattern=${encodeURIComponent(t)}`,n=await fetch(s);if(!n.ok)throw new Error(`indexed-domains failed: ${n.status}`);let i=await n.json();return L.set(t,i),i},W=async(t,e)=>{t.length||(t=[window.location.hostname]);let s=new Set,n=[];for(let i of t)if(i==="*"||z(i))for(let c of await U(i,e))s.has(c.domain)||(s.add(c.domain),n.push(c));else s.has(i)||(s.add(i),n.push({domain:i,page_count:null}));return n},S=new Map,_=()=>{let t=window.location.hostname;return t==="localhost"||t.endsWith(".localhost")||t==="127.0.0.1"},A=t=>_()?`${window.location.origin}/system/semantic-vectors.json?domain=${encodeURIComponent(t)}`:`http://${t}/system/semantic-vectors.json`,H=async t=>{if(S.has(t))return S.get(t);let e=await fetch(A(t));if(!e.ok)return[];let s=await e.json();return S.set(t,s),s},k=async(t,e)=>{let s=await fetch(`${e}/system/embed.json?text=${encodeURIComponent(t)}`);if(!s.ok)throw new Error(`embed failed: ${s.status}`);return(await s.json()).vector},q=async(t,e)=>{let n=(await H(e)).find(i=>i.slug===t);return n?n.vector:null},C=(t,e,{threshold:s,limit:n,excludeSlug:i,excludeDomain:c})=>{let o=[];for(let{domain:a,pages:p}of e)for(let{slug:u,title:r,vector:d}of p){if(u===i&&a===c)continue;let f=0;for(let m=0;m<t.length;m++)f+=t[m]*d[m];f>=s&&o.push({domain:a,slug:u,title:r,score:f})}return o.sort((a,p)=>p.score-a.score),o.slice(0,n)},D=async(t,e)=>{let s=await W(t,e);return(await Promise.all(s.map(async({domain:i})=>({domain:i,pages:await H(i)})))).filter(i=>i.pages.length>0)},T=`
+  .sim-form { display:flex; gap:6px; margin-bottom:8px; }
+  .sim-input { flex:1; padding:6px 8px; font-size:14px; border:1px solid #ccc; border-radius:3px; }
+  .sim-btn { padding:6px 14px; background:#c4561d; color:white; border:none;
+             border-radius:3px; cursor:pointer; font-size:14px; }
+  .sim-btn:disabled { opacity:0.5; cursor:not-allowed; }
+  .sim-status { font-size:12px; color:#888; margin-bottom:6px; min-height:16px; }
+  .sim-results { margin-top:4px; }
+  .sim-result { display:flex; align-items:center; gap:8px; padding:3px 0;
+                border-bottom:1px solid #f0f0f0; }
+  .sim-flag { width:16px; height:16px; vertical-align:middle; margin-right:4px; }
+  .sim-link { font-size:14px; color:#406; flex:1; }
+  .sim-domain { font-size:11px; color:#999; }
+  .similar-results h3 { margin:4px 0 6px; font-size:14px; color:#555; }
+  .similar-results ul { margin:0; padding-left:18px; }
+  .similar-results li { font-size:14px; padding:2px 0; }
+  .similar-results .sim-domain { margin-left:6px; }
+`,O=(t,e)=>`<img class="sim-flag remote" src="${window.wiki.site(t).flag()}"
+        title="${t} \u2014 score ${e.toFixed(3)}"
+        data-site="${t}">`,I=(t,e,s,n)=>`<a class="sim-link" data-title="${s}" data-slug="${e}" data-site="${t}" href="#">${O(t,n)} ${s}</a>`,P=(t,e)=>{let{mode:s,specs:n,threshold:i,limit:c}=R(e?.text||"");if(s==="similar"){let o=n.length?n.join(", "):"current domain";t.html(`
+      <style>${T}</style>
+      <div class="similarity" data-id="${e.id}">
+        <div class="sim-status">Finding similar pages across ${o}\u2026</div>
+        <div class="sim-results"></div>
+      </div>`)}else{let o=n.length?n.join(", "):"(current domain)";t.html(`
+      <style>${T}</style>
+      <div class="similarity" data-id="${e.id}">
+        <div class="sim-form">
+          <input class="sim-input" type="text" placeholder="Search wiki pages\u2026" />
+          <button class="sim-btn">Search</button>
+        </div>
+        <div class="sim-status">Domains: ${o}</div>
+        <div class="sim-results"></div>
+      </div>`)}},V=(t,e)=>{let{mode:s,specs:n,threshold:i,limit:c}=R(e?.text||""),o=window.location.origin,a=t.find(".sim-status")[0],p=t.find(".sim-results")[0];if(t.on("dblclick",()=>window.wiki.textEditor(t,e)),t.on("click",".sim-link",function(u){u.preventDefault();let r=$(this);window.wiki.doInternalLink(r.data("title"),t.parents(".page"),r.data("site"))}),s==="similar")(async()=>{try{let r=t.parents(".page"),d=r.find(".title").text().trim()||document.title,f=F(d),m=window.location.hostname,l=await D(n,o),h=l.reduce((x,w)=>x+w.pages.length,0);a.textContent=`Searching ${h.toLocaleString()} pages\u2026`;let g=await q(f,m);if(!g){a.textContent="Embedding page (not yet indexed)\u2026";let x=r.find(".item").map((w,b)=>$(b).text().trim()).get().filter(Boolean).join(`
+`);g=await k(x||d,o)}let y=C(g,l,{threshold:i,limit:c,excludeSlug:f,excludeDomain:m});if(!y.length){a.textContent=`No similar pages found above threshold ${i}`;return}p.innerHTML=`<h3>Similar Pages</h3><ul>${y.map(({domain:x,slug:w,title:b,score:M})=>`<li>${I(x,w,b,M)}</li>`).join("")}</ul>`,a.textContent=`${y.length} similar pages found`}catch(r){a.textContent=`Error: ${r.message}`}})();else{let u=t.find(".sim-input")[0],r=t.find(".sim-btn")[0],d=null;(async()=>{try{a.textContent="Resolving domains\u2026",d=await D(n,o);let l=d.reduce((h,g)=>h+g.pages.length,0);a.textContent=`Ready \u2014 ${l.toLocaleString()} pages across ${d.length} domains`}catch(l){a.textContent=`Load error: ${l.message}`}})();let m=async()=>{let l=u.value.trim();if(!(!l||!d)){r.disabled=!0,a.textContent="Embedding query\u2026",p.innerHTML="";try{let h=await k(l,o),g=C(h,d,{threshold:0,limit:c,excludeSlug:null,excludeDomain:null});p.innerHTML=g.map(({domain:y,slug:x,title:w,score:b})=>`<div class="sim-result">${I(y,x,w,b)}</div>`).join(""),a.textContent=`Top ${g.length} results for "${l}"`}catch(h){a.textContent=`Error: ${h.message}`}finally{r.disabled=!1}}};r.addEventListener("click",m),u.addEventListener("keydown",l=>{l.key==="Enter"&&m()})}};typeof window<"u"&&(window.plugins=window.plugins||{},window.plugins.similarity={emit:P,bind:V});})();
+//# sourceMappingURL=similarity.js.map

package/client/similarity.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../src/client/similarity.js"],
+  "sourcesContent": ["// wiki-plugin-similarity\n//\n// One item type: similarity\n//\n// DSL in item.text (newline-separated lines):\n//   *                         all indexed domains on this server\n//   david.*                   glob pattern\n//   *.fish                    glob pattern\n//   david.hitchhikers.earth   explicit domain name\n//   # comment                 ignored\n//   SIMILAR: high             threshold preset AND ambient mode trigger\n//   THRESHOLD: 0.72           exact cosine threshold (overrides SIMILAR:)\n//   LIMIT: 8                  max results shown (default 10)\n//\n// Mode is determined by DSL (Ward's ALL-CAPS convention):\n//   SIMILAR: as the FIRST meaningful line \u2192 ambient mode (no search form;\n//     the item automatically finds pages similar to the current page)\n//   Domain specs first, or empty text \u2192 search form mode (user types query)\n//\n// Server endpoints required:\n//   GET  /system/indexed-domains.json?pattern=glob1,glob2\n//   GET  /system/semantic-vectors.json?domain=\n//   GET  /system/embed.json?text=query\n\n// \u2500\u2500 DSL Parser \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nconst SIMILAR_THRESHOLDS = { high: 0.78, medium: 0.68, low: 0.58 }\nconst DEFAULT_THRESHOLD  = SIMILAR_THRESHOLDS.medium\nconst DEFAULT_LIMIT      = 10\n\n// parseDSL returns { mode, specs, threshold, limit }\n//\n// mode: 'similar' if SIMILAR: is the first meaningful line (ambient auto-run),\n//        'search'  otherwise (interactive search form).\n//\n// Ward's convention: ALL-CAPS keyword as first word signals a mode switch.\n// Placing SIMILAR: first makes the item ambient; placing domain specs first\n// (or leaving text empty) keeps it as an interactive search form.\n\nconst parseDSL = text => {\n  const specs   = []\n  let threshold = null\n  let limit     = null\n  let mode      = 'search'  // default: interactive search form\n\n  for (const raw of text.split('\\n')) {\n    const line = raw.trim()\n    if (!line || line.startsWith('#')) continue\n\n    const upper = line.toUpperCase()\n    if (upper.startsWith('SIMILAR:')) {\n      const level = line.split(':')[1].trim().toLowerCase()\n      threshold = SIMILAR_THRESHOLDS[level] ?? DEFAULT_THRESHOLD\n      // SIMILAR: as the first meaningful line \u2192 ambient mode\n      if (!specs.length && mode === 'search') mode = 'similar'\n      continue\n    }\n    if (upper.startsWith('THRESHOLD:')) {\n      threshold = parseFloat(line.split(':')[1]) || DEFAULT_THRESHOLD\n      continue\n    }\n    if (upper.startsWith('LIMIT:')) {\n      limit = parseInt(line.split(':')[1]) || DEFAULT_LIMIT\n      continue\n    }\n    // Anything else is a domain spec\n    specs.push(line)\n  }\n\n  return {\n    mode,\n    specs,\n    threshold: threshold ?? DEFAULT_THRESHOLD,\n    limit:     limit     ?? DEFAULT_LIMIT,\n  }\n}\n\nconst isGlob = spec => spec.includes('*') || spec.includes('?')\n\n// \u2500\u2500 Slug \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nconst slugify = title => title.toLowerCase().replace(/\\s+/g, '-').replace(/[^a-z0-9-]/g, '')\n\n// \u2500\u2500 Domain resolution \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nconst domainCache = new Map()\n\nconst resolveDomainsForSpec = async (spec, origin) => {\n  if (domainCache.has(spec)) return domainCache.get(spec)\n  const url = `${origin}/system/indexed-domains.json?pattern=${encodeURIComponent(spec)}`\n  const res = await fetch(url)\n  if (!res.ok) throw new Error(`indexed-domains failed: ${res.status}`)\n  const list = await res.json()\n  domainCache.set(spec, list)\n  return list\n}\n\nconst resolveDomains = async (specs, origin) => {\n  if (!specs.length) specs = [window.location.hostname]\n  const seen = new Set()\n  const result = []\n  for (const spec of specs) {\n    if (spec === '*' || isGlob(spec)) {\n      for (const item of await resolveDomainsForSpec(spec, origin)) {\n        if (!seen.has(item.domain)) { seen.add(item.domain); result.push(item) }\n      }\n    } else if (!seen.has(spec)) {\n      seen.add(spec); result.push({ domain: spec, page_count: null })\n    }\n  }\n  return result\n}\n\n// \u2500\u2500 Vector loading \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nconst vectorCache = new Map()\n\n// Detect whether the wiki is running on localhost (or a .localhost subdomain).\n// On localhost we proxy all domain vector fetches through the local server's\n// ?domain= parameter \u2014 this reads vectors directly from the farm folder on disk,\n// avoiding any need for remote servers to have this plugin installed, and\n// sidestepping CORS entirely.\nconst isLocalhost = () => {\n  const h = window.location.hostname\n  return h === 'localhost' || h.endsWith('.localhost') || h === '127.0.0.1'\n}\n\nconst vectorUrl = domain => {\n  if (isLocalhost()) {\n    // Route through our local server \u2014 serves any farm domain from disk\n    return `${window.location.origin}/system/semantic-vectors.json?domain=${encodeURIComponent(domain)}`\n  }\n  // On a real server, fetch directly from the remote wiki\n  return `http://${domain}/system/semantic-vectors.json`\n}\n\nconst loadVectors = async domain => {\n  if (vectorCache.has(domain)) return vectorCache.get(domain)\n  const res = await fetch(vectorUrl(domain))\n  if (!res.ok) return []\n  const data = await res.json()\n  vectorCache.set(domain, data)\n  return data\n}\n\n// \u2500\u2500 Embedding \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nconst getEmbedding = async (text, origin) => {\n  const res = await fetch(`${origin}/system/embed.json?text=${encodeURIComponent(text)}`)\n  if (!res.ok) throw new Error(`embed failed: ${res.status}`)\n  return (await res.json()).vector\n}\n\n// Look up an existing page vector from the current domain's cached index.\n// Returns the float[] vector if found, null otherwise.\nconst lookupPageVector = async (slug, domain) => {\n  const pages = await loadVectors(domain)\n  const entry = pages.find(p => p.slug === slug)\n  return entry ? entry.vector : null\n}\n\n// \u2500\u2500 Cosine search \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nconst cosineScan = (queryVec, domainEntries, { threshold, limit, excludeSlug, excludeDomain }) => {\n  const results = []\n  for (const { domain, pages } of domainEntries) {\n    for (const { slug, title, vector } of pages) {\n      if (slug === excludeSlug && domain === excludeDomain) continue\n      let dot = 0\n      for (let i = 0; i < queryVec.length; i++) dot += queryVec[i] * vector[i]\n      if (dot >= threshold) results.push({ domain, slug, title, score: dot })\n    }\n  }\n  results.sort((a, b) => b.score - a.score)\n  return results.slice(0, limit)\n}\n\n// \u2500\u2500 Shared domain loading \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nconst loadDomainEntries = async (specs, origin) => {\n  const domains = await resolveDomains(specs, origin)\n  const entries = await Promise.all(\n    domains.map(async ({ domain }) => ({ domain, pages: await loadVectors(domain) }))\n  )\n  return entries.filter(e => e.pages.length > 0)\n}\n\n// \u2500\u2500 Styles \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nconst STYLES = `\n  .sim-form { display:flex; gap:6px; margin-bottom:8px; }\n  .sim-input { flex:1; padding:6px 8px; font-size:14px; border:1px solid #ccc; border-radius:3px; }\n  .sim-btn { padding:6px 14px; background:#c4561d; color:white; border:none;\n             border-radius:3px; cursor:pointer; font-size:14px; }\n  .sim-btn:disabled { opacity:0.5; cursor:not-allowed; }\n  .sim-status { font-size:12px; color:#888; margin-bottom:6px; min-height:16px; }\n  .sim-results { margin-top:4px; }\n  .sim-result { display:flex; align-items:center; gap:8px; padding:3px 0;\n                border-bottom:1px solid #f0f0f0; }\n  .sim-flag { width:16px; height:16px; vertical-align:middle; margin-right:4px; }\n  .sim-link { font-size:14px; color:#406; flex:1; }\n  .sim-domain { font-size:11px; color:#999; }\n  .similar-results h3 { margin:4px 0 6px; font-size:14px; color:#555; }\n  .similar-results ul { margin:0; padding-left:18px; }\n  .similar-results li { font-size:14px; padding:2px 0; }\n  .similar-results .sim-domain { margin-left:6px; }\n`\n\nconst siteFlag = (domain, score) =>\n  `<img class=\"sim-flag remote\" src=\"${window.wiki.site(domain).flag()}\"\n        title=\"${domain} \u2014 score ${score.toFixed(3)}\"\n        data-site=\"${domain}\">`\n\n// \u2500\u2500 similarity item \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nconst simLink = (domain, slug, title, score) =>\n  `<a class=\"sim-link\" data-title=\"${title}\" data-slug=\"${slug}\" data-site=\"${domain}\" href=\"#\">` +\n  `${siteFlag(domain, score)} ${title}</a>`\n\nexport const emit = (div, item) => {\n  const { mode, specs, threshold, limit } = parseDSL(item?.text || '')\n  if (mode === 'similar') {\n    const label = specs.length ? specs.join(', ') : 'current domain'\n    div.html(`\n      <style>${STYLES}</style>\n      <div class=\"similarity\" data-id=\"${item.id}\">\n        <div class=\"sim-status\">Finding similar pages across ${label}\u2026</div>\n        <div class=\"sim-results\"></div>\n      </div>`)\n  } else {\n    const label = specs.length ? specs.join(', ') : '(current domain)'\n    div.html(`\n      <style>${STYLES}</style>\n      <div class=\"similarity\" data-id=\"${item.id}\">\n        <div class=\"sim-form\">\n          <input class=\"sim-input\" type=\"text\" placeholder=\"Search wiki pages\u2026\" />\n          <button class=\"sim-btn\">Search</button>\n        </div>\n        <div class=\"sim-status\">Domains: ${label}</div>\n        <div class=\"sim-results\"></div>\n      </div>`)\n  }\n}\n\nexport const bind = (div, item) => {\n  const { mode, specs, threshold, limit } = parseDSL(item?.text || '')\n  const origin  = window.location.origin\n  const status  = div.find('.sim-status')[0]\n  const results = div.find('.sim-results')[0]\n\n  div.on('dblclick', () => window.wiki.textEditor(div, item))\n\n  div.on('click', '.sim-link', function (e) {\n    e.preventDefault()\n    const $a = $(this)\n    window.wiki.doInternalLink($a.data('title'), div.parents('.page'), $a.data('site'))\n  })\n\n  if (mode === 'similar') {\n    // Ambient mode \u2014 run automatically, no search form\n    const run = async () => {\n      try {\n        const $page     = div.parents('.page')\n        const pageTitle = $page.find('.title').text().trim() || document.title\n        const currentSlug   = slugify(pageTitle)\n        const currentDomain = window.location.hostname\n\n        const domainEntries = await loadDomainEntries(specs, origin)\n        const total = domainEntries.reduce((n, e) => n + e.pages.length, 0)\n        status.textContent = `Searching ${total.toLocaleString()} pages\u2026`\n\n        let qVec = await lookupPageVector(currentSlug, currentDomain)\n        if (!qVec) {\n          status.textContent = 'Embedding page (not yet indexed)\u2026'\n          const pageText = $page.find('.item').map((_, el) => $(el).text().trim()).get().filter(Boolean).join('\\n')\n          qVec = await getEmbedding(pageText || pageTitle, origin)\n        }\n\n        const scored = cosineScan(qVec, domainEntries, {\n          threshold, limit, excludeSlug: currentSlug, excludeDomain: currentDomain,\n        })\n\n        if (!scored.length) {\n          status.textContent = `No similar pages found above threshold ${threshold}`\n          return\n        }\n        results.innerHTML = `<h3>Similar Pages</h3><ul>${\n          scored.map(({ domain, slug, title, score }) =>\n            `<li>${simLink(domain, slug, title, score)}</li>`).join('')\n        }</ul>`\n        status.textContent = `${scored.length} similar pages found`\n      } catch (e) {\n        status.textContent = `Error: ${e.message}`\n      }\n    }\n    run()\n  } else {\n    // Search form mode \u2014 preload domains, then wait for user input\n    const input = div.find('.sim-input')[0]\n    const btn   = div.find('.sim-btn')[0]\n    let domainEntries = null\n\n    const preload = async () => {\n      try {\n        status.textContent = 'Resolving domains\u2026'\n        domainEntries = await loadDomainEntries(specs, origin)\n        const total = domainEntries.reduce((n, e) => n + e.pages.length, 0)\n        status.textContent = `Ready \u2014 ${total.toLocaleString()} pages across ${domainEntries.length} domains`\n      } catch (e) {\n        status.textContent = `Load error: ${e.message}`\n      }\n    }\n    preload()\n\n    const doSearch = async () => {\n      const query = input.value.trim()\n      if (!query || !domainEntries) return\n      btn.disabled = true\n      status.textContent = 'Embedding query\u2026'\n      results.innerHTML = ''\n      try {\n        const qVec = await getEmbedding(query, origin)\n        const scored = cosineScan(qVec, domainEntries,\n          { threshold: 0, limit, excludeSlug: null, excludeDomain: null })\n        results.innerHTML = scored.map(({ domain, slug, title, score }) =>\n          `<div class=\"sim-result\">${simLink(domain, slug, title, score)}</div>`).join('')\n        status.textContent = `Top ${scored.length} results for \"${query}\"`\n      } catch (e) {\n        status.textContent = `Error: ${e.message}`\n      } finally {\n        btn.disabled = false\n      }\n    }\n\n    btn.addEventListener('click', doSearch)\n    input.addEventListener('keydown', e => { if (e.key === 'Enter') doSearch() })\n  }\n}\n\n// \u2500\u2500 Register item type with the wiki \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nif (typeof window !== 'undefined') {\n  window.plugins = window.plugins || {}\n  window.plugins.similarity = { emit, bind }\n}\n"],
+  "mappings": ";MA0BA,IAAMA,EAAqB,CAAE,KAAM,IAAM,OAAQ,IAAM,IAAK,GAAK,EAC3DC,EAAqBD,EAAmB,OACxCE,EAAqB,GAWrBC,EAAWC,GAAQ,CACvB,IAAMC,EAAU,CAAC,EACbC,EAAY,KACZC,EAAY,KACZC,EAAY,SAEhB,QAAWC,KAAOL,EAAK,MAAM;AAAA,CAAI,EAAG,CAClC,IAAMM,EAAOD,EAAI,KAAK,EACtB,GAAI,CAACC,GAAQA,EAAK,WAAW,GAAG,EAAG,SAEnC,IAAMC,EAAQD,EAAK,YAAY,EAC/B,GAAIC,EAAM,WAAW,UAAU,EAAG,CAChC,IAAMC,EAAQF,EAAK,MAAM,GAAG,EAAE,CAAC,EAAE,KAAK,EAAE,YAAY,EACpDJ,EAAYN,EAAmBY,CAAK,GAAKX,EAErC,CAACI,EAAM,QAAUG,IAAS,WAAUA,EAAO,WAC/C,QACF,CACA,GAAIG,EAAM,WAAW,YAAY,EAAG,CAClCL,EAAY,WAAWI,EAAK,MAAM,GAAG,EAAE,CAAC,CAAC,GAAKT,EAC9C,QACF,CACA,GAAIU,EAAM,WAAW,QAAQ,EAAG,CAC9BJ,EAAQ,SAASG,EAAK,MAAM,GAAG,EAAE,CAAC,CAAC,GAAKR,EACxC,QACF,CAEAG,EAAM,KAAKK,CAAI,CACjB,CAEA,MAAO,CACL,KAAAF,EACA,MAAAH,EACA,UAAWC,GAAaL,EACxB,MAAWM,GAAaL,CAC1B,CACF,EAEMW,EAASC,GAAQA,EAAK,SAAS,GAAG,GAAKA,EAAK,SAAS,GAAG,EAIxDC,EAAUC,GAASA,EAAM,YAAY,EAAE,QAAQ,OAAQ,GAAG,EAAE,QAAQ,cAAe,EAAE,EAIrFC,EAAc,IAAI,IAElBC,EAAwB,MAAOJ,EAAMK,IAAW,CACpD,GAAIF,EAAY,IAAIH,CAAI,EAAG,OAAOG,EAAY,IAAIH,CAAI,EACtD,IAAMM,EAAM,GAAGD,CAAM,wCAAwC,mBAAmBL,CAAI,CAAC,GAC/EO,EAAM,MAAM,MAAMD,CAAG,EAC3B,GAAI,CAACC,EAAI,GAAI,MAAM,IAAI,MAAM,2BAA2BA,EAAI,MAAM,EAAE,EACpE,IAAMC,EAAO,MAAMD,EAAI,KAAK,EAC5B,OAAAJ,EAAY,IAAIH,EAAMQ,CAAI,EACnBA,CACT,EAEMC,EAAiB,MAAOlB,EAAOc,IAAW,CACzCd,EAAM,SAAQA,EAAQ,CAAC,OAAO,SAAS,QAAQ,GACpD,IAAMmB,EAAO,IAAI,IACXC,EAAS,CAAC,EAChB,QAAWX,KAAQT,EACjB,GAAIS,IAAS,KAAOD,EAAOC,CAAI,EAC7B,QAAWY,KAAQ,MAAMR,EAAsBJ,EAAMK,CAAM,EACpDK,EAAK,IAAIE,EAAK,MAAM,IAAKF,EAAK,IAAIE,EAAK,MAAM,EAAGD,EAAO,KAAKC,CAAI,QAE7DF,EAAK,IAAIV,CAAI,IACvBU,EAAK,IAAIV,CAAI,EAAGW,EAAO,KAAK,CAAE,OAAQX,EAAM,WAAY,IAAK,CAAC,GAGlE,OAAOW,CACT,EAIME,EAAc,IAAI,IAOlBC,EAAc,IAAM,CACxB,IAAMC,EAAI,OAAO,SAAS,SAC1B,OAAOA,IAAM,aAAeA,EAAE,SAAS,YAAY,GAAKA,IAAM,WAChE,EAEMC,EAAYC,GACZH,EAAY,EAEP,GAAG,OAAO,SAAS,MAAM,wCAAwC,mBAAmBG,CAAM,CAAC,GAG7F,UAAUA,CAAM,gCAGnBC,EAAc,MAAMD,GAAU,CAClC,GAAIJ,EAAY,IAAII,CAAM,EAAG,OAAOJ,EAAY,IAAII,CAAM,EAC1D,IAAMV,EAAM,MAAM,MAAMS,EAAUC,CAAM,CAAC,EACzC,GAAI,CAACV,EAAI,GAAI,MAAO,CAAC,EACrB,IAAMY,EAAO,MAAMZ,EAAI,KAAK,EAC5B,OAAAM,EAAY,IAAII,EAAQE,CAAI,EACrBA,CACT,EAIMC,EAAe,MAAO9B,EAAMe,IAAW,CAC3C,IAAME,EAAM,MAAM,MAAM,GAAGF,CAAM,2BAA2B,mBAAmBf,CAAI,CAAC,EAAE,EACtF,GAAI,CAACiB,EAAI,GAAI,MAAM,IAAI,MAAM,iBAAiBA,EAAI,MAAM,EAAE,EAC1D,OAAQ,MAAMA,EAAI,KAAK,GAAG,MAC5B,EAIMc,EAAmB,MAAOC,EAAML,IAAW,CAE/C,IAAMM,GADQ,MAAML,EAAYD,CAAM,GAClB,KAAKO,GAAKA,EAAE,OAASF,CAAI,EAC7C,OAAOC,EAAQA,EAAM,OAAS,IAChC,EAIME,EAAa,CAACC,EAAUC,EAAe,CAAE,UAAAnC,EAAW,MAAAC,EAAO,YAAAmC,EAAa,cAAAC,CAAc,IAAM,CAChG,IAAMC,EAAU,CAAC,EACjB,OAAW,CAAE,OAAAb,EAAQ,MAAAc,CAAM,IAAKJ,EAC9B,OAAW,CAAE,KAAAL,EAAM,MAAApB,EAAO,OAAA8B,CAAO,IAAKD,EAAO,CAC3C,GAAIT,IAASM,GAAeX,IAAWY,EAAe,SACtD,IAAII,EAAM,EACV,QAASC,EAAI,EAAGA,EAAIR,EAAS,OAAQQ,IAAKD,GAAOP,EAASQ,CAAC,EAAIF,EAAOE,CAAC,EACnED,GAAOzC,GAAWsC,EAAQ,KAAK,CAAE,OAAAb,EAAQ,KAAAK,EAAM,MAAApB,EAAO,MAAO+B,CAAI,CAAC,CACxE,CAEF,OAAAH,EAAQ,KAAK,CAAC,EAAGK,IAAMA,EAAE,MAAQ,EAAE,KAAK,EACjCL,EAAQ,MAAM,EAAGrC,CAAK,CAC/B,EAIM2C,EAAoB,MAAO7C,EAAOc,IAAW,CACjD,IAAMgC,EAAU,MAAM5B,EAAelB,EAAOc,CAAM,EAIlD,OAHgB,MAAM,QAAQ,IAC5BgC,EAAQ,IAAI,MAAO,CAAE,OAAApB,CAAO,KAAO,CAAE,OAAAA,EAAQ,MAAO,MAAMC,EAAYD,CAAM,CAAE,EAAE,CAClF,GACe,OAAOqB,GAAKA,EAAE,MAAM,OAAS,CAAC,CAC/C,EAIMC,EAAS;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAmBTC,EAAW,CAACvB,EAAQwB,IACxB,qCAAqC,OAAO,KAAK,KAAKxB,CAAM,EAAE,KAAK,CAAC;AAAA,iBACrDA,CAAM,iBAAYwB,EAAM,QAAQ,CAAC,CAAC;AAAA,qBAC9BxB,CAAM,KAIrByB,EAAU,CAACzB,EAAQK,EAAMpB,EAAOuC,IACpC,mCAAmCvC,CAAK,gBAAgBoB,CAAI,gBAAgBL,CAAM,cAC/EuB,EAASvB,EAAQwB,CAAK,CAAC,IAAIvC,CAAK,OAExByC,EAAO,CAACC,EAAKhC,IAAS,CACjC,GAAM,CAAE,KAAAlB,EAAM,MAAAH,EAAO,UAAAC,EAAW,MAAAC,CAAM,EAAIJ,EAASuB,GAAM,MAAQ,EAAE,EACnE,GAAIlB,IAAS,UAAW,CACtB,IAAMmD,EAAQtD,EAAM,OAASA,EAAM,KAAK,IAAI,EAAI,iBAChDqD,EAAI,KAAK;AAAA,eACEL,CAAM;AAAA,yCACoB3B,EAAK,EAAE;AAAA,+DACeiC,CAAK;AAAA;AAAA,aAEvD,CACX,KAAO,CACL,IAAMA,EAAQtD,EAAM,OAASA,EAAM,KAAK,IAAI,EAAI,mBAChDqD,EAAI,KAAK;AAAA,eACEL,CAAM;AAAA,yCACoB3B,EAAK,EAAE;AAAA;AAAA;AAAA;AAAA;AAAA,2CAKLiC,CAAK;AAAA;AAAA,aAEnC,CACX,CACF,EAEaC,EAAO,CAACF,EAAKhC,IAAS,CACjC,GAAM,CAAE,KAAAlB,EAAM,MAAAH,EAAO,UAAAC,EAAW,MAAAC,CAAM,EAAIJ,EAASuB,GAAM,MAAQ,EAAE,EAC7DP,EAAU,OAAO,SAAS,OAC1B0C,EAAUH,EAAI,KAAK,aAAa,EAAE,CAAC,EACnCd,EAAUc,EAAI,KAAK,cAAc,EAAE,CAAC,EAU1C,GARAA,EAAI,GAAG,WAAY,IAAM,OAAO,KAAK,WAAWA,EAAKhC,CAAI,CAAC,EAE1DgC,EAAI,GAAG,QAAS,YAAa,SAAUN,EAAG,CACxCA,EAAE,eAAe,EACjB,IAAMU,EAAK,EAAE,IAAI,EACjB,OAAO,KAAK,eAAeA,EAAG,KAAK,OAAO,EAAGJ,EAAI,QAAQ,OAAO,EAAGI,EAAG,KAAK,MAAM,CAAC,CACpF,CAAC,EAEGtD,IAAS,WAEC,SAAY,CACtB,GAAI,CACF,IAAMuD,EAAYL,EAAI,QAAQ,OAAO,EAC/BM,EAAYD,EAAM,KAAK,QAAQ,EAAE,KAAK,EAAE,KAAK,GAAK,SAAS,MAC3DE,EAAgBlD,EAAQiD,CAAS,EACjCE,EAAgB,OAAO,SAAS,SAEhCzB,EAAgB,MAAMS,EAAkB7C,EAAOc,CAAM,EACrDgD,EAAQ1B,EAAc,OAAO,CAAC2B,EAAGhB,IAAMgB,EAAIhB,EAAE,MAAM,OAAQ,CAAC,EAClES,EAAO,YAAc,aAAaM,EAAM,eAAe,CAAC,eAExD,IAAIE,EAAO,MAAMlC,EAAiB8B,EAAaC,CAAa,EAC5D,GAAI,CAACG,EAAM,CACTR,EAAO,YAAc,yCACrB,IAAMS,EAAWP,EAAM,KAAK,OAAO,EAAE,IAAI,CAACQ,EAAGC,IAAO,EAAEA,CAAE,EAAE,KAAK,EAAE,KAAK,CAAC,EAAE,IAAI,EAAE,OAAO,OAAO,EAAE,KAAK;AAAA,CAAI,EACxGH,EAAO,MAAMnC,EAAaoC,GAAYN,EAAW7C,CAAM,CACzD,CAEA,IAAMsD,EAASlC,EAAW8B,EAAM5B,EAAe,CAC7C,UAAAnC,EAAW,MAAAC,EAAO,YAAa0D,EAAa,cAAeC,CAC7D,CAAC,EAED,GAAI,CAACO,EAAO,OAAQ,CAClBZ,EAAO,YAAc,0CAA0CvD,CAAS,GACxE,MACF,CACAsC,EAAQ,UAAY,6BAClB6B,EAAO,IAAI,CAAC,CAAE,OAAA1C,EAAQ,KAAAK,EAAM,MAAApB,EAAO,MAAAuC,CAAM,IACvC,OAAOC,EAAQzB,EAAQK,EAAMpB,EAAOuC,CAAK,CAAC,OAAO,EAAE,KAAK,EAAE,CAC9D,QACAM,EAAO,YAAc,GAAGY,EAAO,MAAM,sBACvC,OAASrB,EAAG,CACVS,EAAO,YAAc,UAAUT,EAAE,OAAO,EAC1C,CACF,GACI,MACC,CAEL,IAAMsB,EAAQhB,EAAI,KAAK,YAAY,EAAE,CAAC,EAChCiB,EAAQjB,EAAI,KAAK,UAAU,EAAE,CAAC,EAChCjB,EAAgB,MAEJ,SAAY,CAC1B,GAAI,CACFoB,EAAO,YAAc,0BACrBpB,EAAgB,MAAMS,EAAkB7C,EAAOc,CAAM,EACrD,IAAMgD,EAAQ1B,EAAc,OAAO,CAAC2B,EAAGhB,IAAMgB,EAAIhB,EAAE,MAAM,OAAQ,CAAC,EAClES,EAAO,YAAc,gBAAWM,EAAM,eAAe,CAAC,iBAAiB1B,EAAc,MAAM,UAC7F,OAASW,EAAG,CACVS,EAAO,YAAc,eAAeT,EAAE,OAAO,EAC/C,CACF,GACQ,EAER,IAAMwB,EAAW,SAAY,CAC3B,IAAMC,EAAQH,EAAM,MAAM,KAAK,EAC/B,GAAI,GAACG,GAAS,CAACpC,GACf,CAAAkC,EAAI,SAAW,GACfd,EAAO,YAAc,wBACrBjB,EAAQ,UAAY,GACpB,GAAI,CACF,IAAMyB,EAAO,MAAMnC,EAAa2C,EAAO1D,CAAM,EACvCsD,EAASlC,EAAW8B,EAAM5B,EAC9B,CAAE,UAAW,EAAG,MAAAlC,EAAO,YAAa,KAAM,cAAe,IAAK,CAAC,EACjEqC,EAAQ,UAAY6B,EAAO,IAAI,CAAC,CAAE,OAAA1C,EAAQ,KAAAK,EAAM,MAAApB,EAAO,MAAAuC,CAAM,IAC3D,2BAA2BC,EAAQzB,EAAQK,EAAMpB,EAAOuC,CAAK,CAAC,QAAQ,EAAE,KAAK,EAAE,EACjFM,EAAO,YAAc,OAAOY,EAAO,MAAM,iBAAiBI,CAAK,GACjE,OAASzB,EAAG,CACVS,EAAO,YAAc,UAAUT,EAAE,OAAO,EAC1C,QAAE,CACAuB,EAAI,SAAW,EACjB,EACF,EAEAA,EAAI,iBAAiB,QAASC,CAAQ,EACtCF,EAAM,iBAAiB,UAAWtB,GAAK,CAAMA,EAAE,MAAQ,SAASwB,EAAS,CAAE,CAAC,CAC9E,CACF,EAII,OAAO,OAAW,MACpB,OAAO,QAAU,OAAO,SAAW,CAAC,EACpC,OAAO,QAAQ,WAAa,CAAE,KAAAnB,EAAM,KAAAG,CAAK",
+  "names": ["SIMILAR_THRESHOLDS", "DEFAULT_THRESHOLD", "DEFAULT_LIMIT", "parseDSL", "text", "specs", "threshold", "limit", "mode", "raw", "line", "upper", "level", "isGlob", "spec", "slugify", "title", "domainCache", "resolveDomainsForSpec", "origin", "url", "res", "list", "resolveDomains", "seen", "result", "item", "vectorCache", "isLocalhost", "h", "vectorUrl", "domain", "loadVectors", "data", "getEmbedding", "lookupPageVector", "slug", "entry", "p", "cosineScan", "queryVec", "domainEntries", "excludeSlug", "excludeDomain", "results", "pages", "vector", "dot", "i", "b", "loadDomainEntries", "domains", "e", "STYLES", "siteFlag", "score", "simLink", "emit", "div", "label", "bind", "status", "$a", "$page", "pageTitle", "currentSlug", "currentDomain", "total", "n", "qVec", "pageText", "_", "el", "scored", "input", "btn", "doSearch", "query"]
+}

package/factory.json

@@ -0,0 +1 @@
+{"name": "Similarity", "title": "Semantic similarity search across wiki domains", "category": "other"}

package/package.json

@@ -0,0 +1,29 @@
+{
+  "name": "wiki-plugin-similarity",
+  "version": "0.1.0",
+  "description": "Federated Wiki plugin — semantic similarity search across wiki domains using prebuilt vector indexes",
+  "keywords": ["wiki", "federated wiki", "plugin", "similarity", "semantic search", "embeddings"],
+  "author": {
+    "name": "David Bovill",
+    "url": "https://github.com/Hitchhikers-Guide-to-the-Galaxy"
+  },
+  "license": "MIT",
+  "type": "module",
+  "files": ["client", "server", "pages", "factory.json"],
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/Hitchhikers-Guide-to-the-Galaxy/wiki-plugin-similarity.git"
+  },
+  "engines": { "node": ">=18.x" },
+  "scripts": {
+    "build": "npm run clean && npm run test && node --no-warnings scripts/build-client.js",
+    "dev":   "node --no-warnings scripts/build-client.js --watch",
+    "about": "wiki -p 3010 -d . --security_legacy",
+    "clean": "rm -f client/similarity.js client/similarity.js.map",
+    "test":  "node --test",
+    "prepublishOnly": "npm run build"
+  },
+  "devDependencies": {
+    "esbuild": "^0.25.0"
+  }
+}

package/pages/about-similarity-plugin

@@ -0,0 +1,53 @@
+{
+  "title": "About Similarity Plugin",
+  "story": [
+    {
+      "type": "markdown",
+      "id": "61ccb2d2cdb60c71",
+      "text": "The Similarity plugin adds semantic similarity search to a Federated Wiki farm. It ships two item types — **similarity** (an interactive search form) and **similar** (an ambient panel that automatically finds pages related to the one you are reading)."
+    },
+    {
+      "type": "similarity",
+      "id": "ad3c41667c285efd",
+      "text": "SIMILAR: medium\nLIMIT: 5"
+    },
+    {
+      "type": "markdown",
+      "id": "bdf93e3860d64ebb",
+      "text": "## Domain DSL"
+    },
+    {
+      "type": "markdown",
+      "id": "0d628d43edf9f8e8",
+      "text": "The item's `text` field configures which domains to search and how similar results must be:\n\n```\n*                        all indexed domains on this server\ndavid.*                  glob pattern\ndavid.hitchhikers.earth  explicit domain\n# comment                ignored\n---\nSIMILAR: high            threshold preset  (high=0.78 / medium=0.68 / low=0.58)\nTHRESHOLD: 0.72          exact cosine threshold\nLIMIT: 8                 max results (default 10)\n```\n\nLeave `text` empty to search the current domain at the default medium threshold."
+    },
+    {
+      "type": "markdown",
+      "id": "c0775c94a895e9d2",
+      "text": "## How It Works"
+    },
+    {
+      "type": "markdown",
+      "id": "4b1bbe95544548eb",
+      "text": "Pages are pre-indexed as 384-dimensional embedding vectors using BAAI/bge-small-en-v1.5 via fastembed. The plugin loads `semantic-vectors.json` index files directly in the browser and scores them with cosine similarity in JavaScript — no remote search engine required.\n\nThe **similar** item type uses a fast path: if the current page is already indexed, its vector is reused directly without an embedding call. See [[Similarity Indexing]] for how to build and refresh indices."
+    }
+  ],
+  "journal": [
+    {
+      "type": "create",
+      "item": {
+        "title": "About Similarity Plugin",
+        "story": [
+          {
+            "type": "markdown",
+            "id": "61ccb2d2cdb60c71",
+            "text": "The Similarity plugin adds semantic similarity search to a Federated Wiki farm. It ships two item types — **similarity** (an interactive search form) and **similar** (an ambient panel that automatically finds pages related to the one you are reading)."
+          }
+        ]
+      },
+      "date": 1781026832000,
+      "certificate": "from marvin"
+    }
+  ],
+  "plugin": "similarity"
+}

package/server/server.js

@@ -0,0 +1,198 @@
+// wiki-plugin-similarity — server-side component
+//
+// Registers three routes with the wiki's Express app via the startServer(params)
+// hook in wiki-server/lib/plugins.js.
+//
+// Routes:
+//   GET /system/indexed-domains.json?pattern=glob1,glob2
+//     Returns [{domain, page_count}] for all domains whose semantic-vectors.json
+//     exists and whose name matches any of the supplied glob patterns.
+//
+//   GET /system/semantic-vectors.json
+//     Serves {farm}/{req.hostname}/status/semantic-vectors.json.
+//     req.hostname selects the domain in a fedwiki farm.
+//
+//   GET /system/embed?text=…
+//     Proxies to the local FastAPI /embed endpoint and returns {vector:[…]}.
+//     Requires the FastAPI server to be running on EMBED_URL (default localhost:8000).
+//
+// Farm root is derived from argv.status which is {farm}/{domain}/status.
+
+import fs   from 'node:fs'
+import path from 'node:path'
+import http from 'node:http'
+
+const EMBED_URL      = process.env.WIKI_EMBED_URL   || 'http://localhost:8000/embed'
+// Optional additional farm roots to search when a domain's vectors aren't found
+// in the primary farm. Colon-separated list of absolute paths, e.g.:
+//   WIKI_EXTRA_FARMS=/Users/david/Nextcloud/fedwiki:/Users/david/other-farm
+const EXTRA_FARMS    = (process.env.WIKI_EXTRA_FARMS || '').split(':').filter(Boolean)
+
+// ── Glob matching ─────────────────────────────────────────────────────────────
+// Supports * (any chars) and ? (one char). No path separator semantics needed.
+
+const globMatch = (pattern, str) => {
+  const p = pattern.length
+  const s = str.length
+  const dp = Array.from({ length: p + 1 }, () => new Array(s + 1).fill(false))
+  dp[0][0] = true
+  for (let i = 1; i <= p; i++) {
+    if (pattern[i - 1] === '*') dp[i][0] = dp[i - 1][0]
+  }
+  for (let i = 1; i <= p; i++) {
+    for (let j = 1; j <= s; j++) {
+      if (pattern[i - 1] === '*') {
+        dp[i][j] = dp[i - 1][j] || dp[i][j - 1]
+      } else if (pattern[i - 1] === '?' || pattern[i - 1] === str[j - 1]) {
+        dp[i][j] = dp[i - 1][j - 1]
+      }
+    }
+  }
+  return dp[p][s]
+}
+
+const matchesAny = (domain, patterns) =>
+  patterns.some(p => p === '*' || globMatch(p, domain))
+
+// ── Multi-farm helpers ────────────────────────────────────────────────────────
+
+// Return the first path that exists across all farm roots for a given domain
+// and relative sub-path (e.g. 'status/semantic-vectors.json').
+const findInFarms = (farmRoot, domain, relPath) => {
+  const farms = [farmRoot, ...EXTRA_FARMS]
+  for (const farm of farms) {
+    const full = path.join(farm, domain, relPath)
+    try { fs.accessSync(full, fs.constants.F_OK); return full } catch { /* try next */ }
+  }
+  return null
+}
+
+// ── Discover indexed domains ───────────────────────────────────────────────────
+
+const findIndexedDomains = (farmRoot, patterns) => {
+  const farms = [farmRoot, ...EXTRA_FARMS]
+  const seen  = new Set()
+  const results = []
+
+  for (const farm of farms) {
+    let entries
+    try { entries = fs.readdirSync(farm, { withFileTypes: true }) } catch { continue }
+
+    for (const ent of entries) {
+      if (!ent.isDirectory()) continue
+      const domain = ent.name
+      if (seen.has(domain)) continue          // first farm wins
+      if (!matchesAny(domain, patterns)) continue
+      const vecFile = path.join(farm, domain, 'status', 'semantic-vectors.json')
+      try { fs.accessSync(vecFile, fs.constants.F_OK) } catch { continue }
+      seen.add(domain)
+      let pageCount = null
+      try {
+        const pages = JSON.parse(fs.readFileSync(vecFile, 'utf8'))
+        pageCount = Array.isArray(pages) ? pages.length : null
+      } catch { /* ignore */ }
+      results.push({ domain, page_count: pageCount })
+    }
+  }
+
+  results.sort((a, b) => a.domain.localeCompare(b.domain))
+  return results
+}
+
+// ── Fetch helper (node:http, avoids fetch() version concerns) ─────────────────
+
+const postJson = (url, body) =>
+  new Promise((resolve, reject) => {
+    const payload = JSON.stringify(body)
+    const u = new URL(url)
+    const opts = {
+      hostname: u.hostname,
+      port:     u.port || 80,
+      path:     u.pathname + u.search,
+      method:   'POST',
+      headers:  {
+        'Content-Type':   'application/json',
+        'Content-Length': Buffer.byteLength(payload),
+      },
+    }
+    const req = http.request(opts, res => {
+      let data = ''
+      res.on('data', chunk => { data += chunk })
+      res.on('end', () => {
+        try { resolve(JSON.parse(data)) }
+        catch (e) { reject(new Error(`embed response parse error: ${e.message}`)) }
+      })
+    })
+    req.on('error', reject)
+    req.write(payload)
+    req.end()
+  })
+
+// ── startServer — called by wiki-server/lib/plugins.js ────────────────────────
+
+export const startServer = ({ argv, app }) => {
+  // Farm root: argv.status = {farm}/{thisDomain}/status  →  go up two levels
+  const farmRoot = path.dirname(path.dirname(argv.status))
+
+  console.log('[wiki-plugin-similarity] registering /system routes, farm:', farmRoot)
+
+  // CORS helper
+  const cors = res => {
+    res.setHeader('Access-Control-Allow-Origin', '*')
+    res.setHeader('Access-Control-Allow-Methods', 'GET, OPTIONS')
+    res.setHeader('Access-Control-Allow-Headers', 'Content-Type')
+  }
+
+  // ── OPTIONS pre-flight ─────────────────────────────────────────────────────
+  app.options('/system/indexed-domains.json', (req, res) => {
+    cors(res); res.sendStatus(204)
+  })
+  app.options('/system/semantic-vectors.json', (req, res) => {
+    cors(res); res.sendStatus(204)
+  })
+  app.options('/system/embed.json', (req, res) => {
+    cors(res); res.sendStatus(204)
+  })
+
+  // ── GET /system/indexed-domains.json?pattern=glob1,glob2 ──────────────────
+  app.get('/system/indexed-domains.json', (req, res) => {
+    cors(res)
+    const raw = req.query.pattern || '*'
+    const patterns = raw.split(',').map(p => p.trim()).filter(Boolean)
+    const results = findIndexedDomains(farmRoot, patterns)
+    res.json(results)
+  })
+
+  // ── GET /system/semantic-vectors.json ─────────────────────────────────────
+  // Two modes:
+  //   ?domain=david.hitchhikers.earth  — local proxy: searches all farm roots
+  //       on disk (private farm + WIKI_EXTRA_FARMS). Lets localhost clients load
+  //       vectors for any domain without CORS or remote plugin requirements.
+  //   (no ?domain)  — req.hostname selects the domain; works for farm requests
+  //       where Caddy routes each hostname to this server.
+  app.get('/system/semantic-vectors.json', (req, res) => {
+    cors(res)
+    const domain  = req.query.domain || req.hostname || 'localhost'
+    const vecFile = findInFarms(farmRoot, domain, 'status/semantic-vectors.json')
+    if (!vecFile) {
+      return res.status(404).json({ error: `vectors not found for ${domain}` })
+    }
+    res.sendFile(vecFile)
+  })
+
+  // ── GET /system/embed.json?text=… ────────────────────────────────────────
+  app.get('/system/embed.json', async (req, res) => {
+    cors(res)
+    const text = req.query.text
+    if (!text) return res.status(400).json({ error: 'text parameter required' })
+    try {
+      const result = await postJson(EMBED_URL, { text })
+      res.json(result)
+    } catch (e) {
+      console.error('[wiki-plugin-similarity] embed proxy error:', e.message)
+      res.status(502).json({ error: `embed service unavailable: ${e.message}` })
+    }
+  })
+
+  console.log('[wiki-plugin-similarity] routes registered')
+}