{"id":50052,"date":"2026-09-04T22:21:04","date_gmt":"2026-09-04T22:21:04","guid":{"rendered":"https:\/\/parashift.ai\/?page_id=50052"},"modified":"2026-09-05T15:04:10","modified_gmt":"2026-09-05T15:04:10","slug":"dokumentli","status":"publish","type":"page","link":"https:\/\/parashift.ai\/de\/dokumentli\/","title":{"rendered":"Dokumentli\u00ae \u2013 Das spezialisierte Vision-Language-Model f\u00fcr Dokumenten-Intelligence"},"content":{"rendered":"\t\t<div data-elementor-type=\"wp-page\" data-elementor-id=\"50052\" class=\"elementor elementor-50052\" data-elementor-post-type=\"page\">\n\t\t\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-363ea10 elementor-section-full_width elementor-section-stretched elementor-section-height-default elementor-section-height-default\" data-id=\"363ea10\" data-element_type=\"section\" data-e-type=\"section\" data-settings=\"{&quot;stretch_section&quot;:&quot;section-stretched&quot;,&quot;pix_scale_in&quot;:&quot;none&quot;}\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-d87dce3\" data-id=\"d87dce3\" data-element_type=\"column\" data-e-type=\"column\" data-settings=\"{&quot;pix_scale_in&quot;:&quot;none&quot;}\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t<div class=\"elementor-element elementor-element-7debc98 elementor-widget elementor-widget-html\" data-id=\"7debc98\" data-element_type=\"widget\" data-e-type=\"widget\" data-widget_type=\"html.default\">\n\t\t\t\t\t<link rel=\"preconnect\" href=\"https:\/\/fonts.googleapis.com\">\n<link rel=\"preconnect\" href=\"https:\/\/fonts.gstatic.com\" crossorigin>\n<link href=\"https:\/\/fonts.googleapis.com\/css2?family=Barlow:wght@400;500;600;700;800&family=JetBrains+Mono:wght@400;500;600&display=swap\" rel=\"stylesheet\">\n<style>\n.psvr-scope{\n  --bg:#ffffff;--bg-alt:#f3f6fc;--bg-deep:#0b1526;--bg-deep-alt:#0f1c33;\n  --ink:#16202c;--ink-soft:#55626f;--ink-faint:#8792a0;\n  --invert:#eef2fa;--invert-soft:#9fadc4;\n  --brand:#4987f5;--brand-deep:#2f66d6;--brand-glow:#e8f0fe;--teal:#7dabf9;--good:#1f9d6f;\n  --line:#e2e8f2;--line-deep:#233047;--radius:10px;\n  --shadow:0 1px 2px rgba(20,30,50,.04), 0 8px 24px -12px rgba(20,30,50,.10);\n  background:var(--bg);color:var(--ink);font-family:\"Barlow\",Arial,Helvetica,sans-serif;\n  line-height:1.55;-webkit-font-smoothing:antialiased;display:block;overflow-x:hidden;\n  margin:-10px 0;\n  margin-left:calc(50% - 50vw);margin-right:calc(50% - 50vw);width:100vw;\n}\n.psvr-scope, .psvr-scope *{box-sizing:border-box;}\n.psvr-scope img{max-width:100%;display:block;}\n.psvr-scope a{color:inherit;text-decoration:none;}\n.psvr-scope h1,.psvr-scope h2,.psvr-scope h3,.psvr-scope h4{font-family:\"Barlow\",Arial,Helvetica,sans-serif;line-height:1.12;text-wrap:balance;margin:0;font-weight:800;letter-spacing:-.01em;}\n.psvr-scope p{margin:0;}\n.psvr-scope .mono{font-family:\"JetBrains Mono\",ui-monospace,Consolas,monospace;font-variant-numeric:tabular-nums;}\n.psvr-scope .wrap{max-width:1180px;margin:0 auto;padding:0 32px;}\n@media (max-width:720px){.psvr-scope .wrap{padding:0 20px;}}\n.psvr-scope .btn{display:inline-flex;align-items:center;gap:8px;background:var(--brand);color:#fff;padding:13px 24px;border-radius:8px;font-weight:700;font-size:15px;border:1px solid var(--brand);cursor:pointer;transition:transform .12s ease, background .15s ease, box-shadow .15s ease;box-shadow:0 6px 16px -8px rgba(73,135,245,.55);}\n.psvr-scope .btn:hover{background:var(--brand-deep);border-color:var(--brand-deep);transform:translateY(-1px);color:#fff;}\n.psvr-scope .btn.ghost{background:transparent;color:var(--brand);border:1px solid var(--line);box-shadow:none;}\n.psvr-scope .btn.ghost:hover{background:var(--brand-glow);transform:none;color:var(--brand);}\n.psvr-scope .eyebrow{display:inline-flex;align-items:center;gap:8px;font-family:\"JetBrains Mono\",monospace;font-size:11.5px;font-weight:600;letter-spacing:.06em;text-transform:uppercase;color:var(--brand);background:var(--brand-glow);border:1px solid rgba(73,135,245,.25);padding:6px 12px;border-radius:100px;}\n.psvr-scope .cite{display:inline-flex;align-items:center;gap:6px;font-family:\"JetBrains Mono\",monospace;font-size:11.5px;font-weight:500;color:var(--ink-soft);background:var(--bg-alt);border:1px solid var(--line);padding:3px 9px;border-radius:5px;white-space:nowrap;}\n.psvr-scope .cite b{color:var(--ink);font-weight:600;}\n.psvr-scope section{padding:96px 0;margin-left:calc(50% - 50vw);margin-right:calc(50% - 50vw);width:100vw;}\n@media (max-width:720px){.psvr-scope section{padding:64px 0;}}\n.psvr-scope .section-head{max-width:660px;margin:0 0 48px;}\n.psvr-scope .intchain{margin-top:24px;font-family:\"JetBrains Mono\",monospace;font-size:13px;letter-spacing:.02em;color:var(--ink-faint);}\n.psvr-scope .section-head .eyebrow{margin-bottom:16px;}\n.psvr-scope .section-head h2{font-size:clamp(28px,3.4vw,38px);}\n.psvr-scope .section-head p{margin-top:12px;color:var(--ink-soft);font-size:16px;max-width:60ch;}\n.psvr-scope .alt{background:var(--bg-alt);margin-left:calc(50% - 50vw);margin-right:calc(50% - 50vw);width:100vw;}\n.psvr-scope .deep{background:var(--bg-deep);color:var(--invert);margin-left:calc(50% - 50vw);margin-right:calc(50% - 50vw);width:100vw;}\n.psvr-scope .deep h2,.psvr-scope .deep h3{color:var(--invert);}\n.psvr-scope .deep .section-head p{color:var(--invert-soft);}\n\n\/* hero \u2014 background matches parashift.ai homepage hero exactly *\/\n.psvr-scope .hero{background:linear-gradient(160deg,#030303 0%,#838890 100%);color:var(--invert);padding:168px 0 96px;position:relative;overflow:hidden;margin-left:calc(50% - 50vw);margin-right:calc(50% - 50vw);width:100vw;}\n.psvr-scope .hero .wrap{position:relative;z-index:1;}\n.psvr-scope .hero .eyebrow{background:rgba(73,135,245,.18);color:#9db8f7;border-color:rgba(73,135,245,.35);}\n.psvr-scope .hero-grid{display:grid;grid-template-columns:1.15fr 1fr;gap:64px;align-items:center;}\n@media (max-width:880px){.psvr-scope .hero-grid{grid-template-columns:1fr;}}\n.psvr-scope .hero h1{font-size:clamp(32px,4vw,46px);color:var(--invert);margin:18px 0 20px;}\n.psvr-scope .hero .lede{font-size:18px;color:var(--invert-soft);max-width:56ch;margin-bottom:28px;}\n.psvr-scope .hero .ctas{display:flex;gap:14px;flex-wrap:wrap;margin-bottom:32px;}\n.psvr-scope .hero .btn.ghost{color:var(--invert);border-color:var(--line-deep);}\n.psvr-scope .hero .btn.ghost:hover{background:rgba(255,255,255,.07);}\n.psvr-scope .hero .microproof{display:flex;gap:22px;flex-wrap:wrap;font-size:13px;color:var(--invert-soft);}\n.psvr-scope .hero .microproof span{display:flex;align-items:center;gap:7px;}\n.psvr-scope .hero .microproof .dot{width:6px;height:6px;border-radius:50%;background:var(--good);}\n\n\/* ai-aura \u2014 deep-space animated backdrop: nebula bloom, starfield, radar sweep, pulse rings *\/\n.psvr-scope .ai-aura{position:relative;isolation:isolate;}\n.psvr-scope .ai-aura > span{position:absolute;z-index:0;pointer-events:none;display:block;}\n.psvr-scope .ai-aura .aura-vignette{inset:-110px;border-radius:50%;\n  background:radial-gradient(circle at 50% 42%, rgba(2,4,14,.88) 0%, rgba(3,3,12,.6) 40%, rgba(3,3,12,0) 72%);\n}\n.psvr-scope .ai-aura .aura-stars{inset:-140px;opacity:.9;\n  background-image:\n    radial-gradient(1.6px 1.6px at 12% 22%, #fff 60%, transparent),\n    radial-gradient(1.2px 1.2px at 28% 68%, #cfe4ff 60%, transparent),\n    radial-gradient(1.8px 1.8px at 44% 14%, #fff 60%, transparent),\n    radial-gradient(1.1px 1.1px at 61% 78%, #bfe6ff 60%, transparent),\n    radial-gradient(1.5px 1.5px at 76% 32%, #fff 60%, transparent),\n    radial-gradient(1.2px 1.2px at 85% 60%, #cfe4ff 60%, transparent),\n    radial-gradient(1.7px 1.7px at 8% 82%, #fff 60%, transparent),\n    radial-gradient(1.1px 1.1px at 52% 46%, #fff 60%, transparent),\n    radial-gradient(1.4px 1.4px at 93% 18%, #bfe6ff 60%, transparent),\n    radial-gradient(1.1px 1.1px at 35% 92%, #fff 60%, transparent);\n  background-repeat:no-repeat;\n  animation:ai-twinkle 4s ease-in-out infinite alternate;\n}\n.psvr-scope .ai-aura .aura-stars.s2{inset:-100px;opacity:.5;transform:scale(1.3);\n  animation:ai-twinkle 5.5s ease-in-out infinite alternate-reverse;\n}\n.psvr-scope .ai-aura .aura-glow{inset:-130px;filter:blur(40px);mix-blend-mode:screen;\n  background:\n    radial-gradient(circle at 20% 18%, rgba(96,155,255,.14), transparent 52%),\n    radial-gradient(circle at 88% 12%, rgba(94,224,255,.11), transparent 48%),\n    radial-gradient(circle at 66% 96%, rgba(199,110,255,.10), transparent 55%),\n    radial-gradient(circle at 6% 88%, rgba(120,90,255,.08), transparent 50%);\n  animation:ai-drift 15s ease-in-out infinite alternate;\n}\n.psvr-scope .ai-aura .aura-nebula2{inset:-160px;filter:blur(60px);mix-blend-mode:screen;opacity:.65;\n  background:\n    radial-gradient(ellipse 60% 40% at 30% 70%, rgba(224,94,255,.35), transparent 60%),\n    radial-gradient(ellipse 50% 35% at 78% 30%, rgba(94,180,255,.35), transparent 60%);\n  animation:ai-drift2 22s ease-in-out infinite alternate;\n}\n.psvr-scope .ai-aura .aura-corona{top:50%;left:50%;width:260px;height:260px;border-radius:50%;transform:translate(-50%,-50%) scale(1);\n  background:radial-gradient(circle, rgba(190,230,255,.55) 0%, rgba(120,150,255,.28) 38%, transparent 72%);\n  filter:blur(6px);mix-blend-mode:screen;\n  animation:ai-corona-breathe 3.4s ease-in-out infinite;\n}\n.psvr-scope .ai-aura .aura-grid{inset:-70px;\n  background-image:radial-gradient(rgba(255,255,255,.4) 1.3px, transparent 1.8px);\n  background-size:27px 27px;\n  opacity:.45;\n  -webkit-mask-image:radial-gradient(circle at 50% 44%, black 0%, transparent 66%);\n  mask-image:radial-gradient(circle at 50% 44%, black 0%, transparent 66%);\n  animation:ai-grid-pan 26s linear infinite;\n}\n.psvr-scope .ai-aura .aura-radar{display:none;}\n.psvr-scope .ai-aura .pulse-ring{top:50%;left:50%;width:280px;height:280px;border:1.5px solid rgba(120,220,255,.7);border-radius:50%;transform:translate(-50%,-50%) scale(.55);opacity:0;animation:ai-pulse 5s ease-out infinite;}\n.psvr-scope .ai-aura .pulse-ring.r2{animation-delay:1.7s;border-color:rgba(224,94,255,.55);}\n.psvr-scope .ai-aura .pulse-ring.r3{animation-delay:3.3s;}\n.psvr-scope .ai-aura > .model-visual{position:relative;z-index:1;}\n@keyframes ai-drift{0%{transform:translate(0,0) scale(1);}50%{transform:translate(18px,-14px) scale(1.1);}100%{transform:translate(-14px,10px) scale(1);}}\n@keyframes ai-drift2{0%{transform:translate(0,0) scale(1);}50%{transform:translate(-20px,16px) scale(1.14);}100%{transform:translate(16px,-12px) scale(1);}}\n@keyframes ai-twinkle{0%,100%{opacity:.5;}50%{opacity:1;}}\n@keyframes ai-corona-breathe{0%,100%{opacity:.55;transform:translate(-50%,-50%) scale(.92);}50%{opacity:.95;transform:translate(-50%,-50%) scale(1.12);}}\n@keyframes ai-grid-pan{0%{background-position:0 0;}100%{background-position:54px 54px;}}\n@keyframes ai-radar-spin{to{transform:translate(-50%,-50%) rotate(360deg);}}\n@keyframes ai-pulse{0%{transform:translate(-50%,-50%) scale(.5);opacity:.7;}75%{opacity:.1;}100%{transform:translate(-50%,-50%) scale(1.35);opacity:0;}}\n@media (prefers-reduced-motion: reduce){.psvr-scope .ai-aura .aura-glow,.psvr-scope .ai-aura .aura-nebula2,.psvr-scope .ai-aura .aura-corona,.psvr-scope .ai-aura .aura-grid,.psvr-scope .ai-aura .aura-radar,.psvr-scope .ai-aura .aura-stars,.psvr-scope .ai-aura .pulse-ring{animation:none;}.psvr-scope .ai-aura .pulse-ring{display:none;}}\n.psvr-scope .model-visual{position:relative;}\n.psvr-scope .mv-svg{width:100%;height:auto;display:block;overflow:visible;}\n.psvr-scope .mv-path{fill:none;stroke:url(#flowLine);stroke-width:2;stroke-linecap:round;stroke-dasharray:5 9;animation:mv-flow .9s linear infinite;}\n.psvr-scope .mv-path.out{animation-duration:.75s;}\n.psvr-scope .mv-path-glow{fill:none;stroke:url(#flowLine);stroke-width:6;stroke-linecap:round;opacity:.3;filter:blur(3.5px);}\n@keyframes mv-flow{to{stroke-dashoffset:-28;}}\n.psvr-scope .mv-doc-anim{animation:mv-float 2.6s cubic-bezier(.45,0,.55,1) infinite;transform-origin:center;transform-box:fill-box;}\n.psvr-scope .mv-doc-anim.d1{animation-delay:0s;}\n.psvr-scope .mv-doc-anim.d2{animation-delay:.85s;}\n.psvr-scope .mv-doc-anim.d3{animation-delay:1.7s;}\n@keyframes mv-float{0%,100%{transform:translate(0,0) scale(1);opacity:.94;}50%{transform:translate(5px,-9px) scale(1.035);opacity:1;}}\n.psvr-scope .mv-scanline{animation:mv-scan 2.6s cubic-bezier(.45,0,.55,1) infinite;}\n.psvr-scope .mv-doc-anim.d2 .mv-scanline{animation-delay:.85s;}\n.psvr-scope .mv-doc-anim.d3 .mv-scanline{animation-delay:1.7s;}\n@keyframes mv-scan{0%{transform:translateY(-4px);opacity:0;}10%{opacity:1;}50%{opacity:.9;}62%{opacity:0;}100%{transform:translateY(96px);opacity:0;}}\n.psvr-scope .mv-core-pulse{animation:mv-corepulse 2.1s ease-in-out infinite;transform-origin:320px 280px;}\n@keyframes mv-corepulse{0%,100%{opacity:.58;transform:scale(1);}50%{opacity:1;transform:scale(1.1);}}\n.psvr-scope .mv-core-facets{transform-origin:320px 280px;animation:mv-facet-spin 24s linear infinite;}\n.psvr-scope .mv-core-facets.rev{animation-direction:reverse;animation-duration:17s;}\n@keyframes mv-facet-spin{to{transform:rotate(360deg);}}\n.psvr-scope .mv-node{filter:drop-shadow(0 0 5px rgba(94,224,255,.95));}\n.psvr-scope .mv-spark{animation:mv-spark-twinkle 2.4s ease-in-out infinite;}\n@keyframes mv-spark-twinkle{0%,100%{opacity:.15;}50%{opacity:.95;}}\n.psvr-scope .mv-stream-doc{filter:drop-shadow(0 3px 6px rgba(10,20,40,.28));}\n.psvr-scope .mv-stream{offset-rotate:0deg;animation-name:mv-stream-move;animation-timing-function:linear;animation-iteration-count:infinite;}\n.psvr-scope .mv-stream.p-top{offset-path:path(\"M118,132 C230,140 246,224 318,272\");animation-duration:2.6s;}\n.psvr-scope .mv-stream.p-mid{offset-path:path(\"M100,296 C220,300 252,290 318,278\");animation-duration:2.3s;}\n.psvr-scope .mv-stream.p-bot{offset-path:path(\"M120,458 C236,452 258,344 318,286\");animation-duration:3s;}\n.psvr-scope .mv-stream.p-out{offset-path:path(\"M322,278 C398,232 462,208 538,150\");animation-name:mv-stream-move-out;animation-duration:1.9s;}\n@keyframes mv-stream-move{0%{offset-distance:0%;opacity:0;}12%{opacity:1;}82%{opacity:1;}100%{offset-distance:100%;opacity:0;}}\n@keyframes mv-stream-move-out{0%{offset-distance:0%;opacity:0;}10%{opacity:1;}80%{opacity:1;}100%{offset-distance:100%;opacity:0;}}\n.psvr-scope .mv-stream.t0{animation-delay:0s;}\n.psvr-scope .mv-stream.t1{animation-delay:-.3s;}\n.psvr-scope .mv-stream.t2{animation-delay:-.5s;}\n.psvr-scope .mv-stream.t3{animation-delay:-.75s;}\n.psvr-scope .mv-stream.t4{animation-delay:-.9s;}\n.psvr-scope .mv-stream.t5{animation-delay:-1s;}\n.psvr-scope .mv-stream.t6{animation-delay:-1.15s;}\n.psvr-scope .mv-stream.t7{animation-delay:-1.5s;}\n.psvr-scope .mv-stream.t8{animation-delay:-1.7s;}\n.psvr-scope .mv-stream.t9{animation-delay:-2s;}\n.psvr-scope .mv-stream.t10{animation-delay:-2.1s;}\n.psvr-scope .mv-stream.t11{animation-delay:-2.3s;}\n.psvr-scope .mv-json-row{animation:mv-row-fade 3.2s ease-in-out infinite;}\n.psvr-scope .mv-json-row.r1{animation-delay:0s;}\n.psvr-scope .mv-json-row.r2{animation-delay:.3s;}\n.psvr-scope .mv-json-row.r3{animation-delay:.6s;}\n@keyframes mv-row-fade{0%,100%{opacity:.45;}15%,45%{opacity:1;}}\n.psvr-scope .mv-json-check{animation:mv-check-pulse 2.4s ease-in-out infinite;transform-origin:610px 182px;}\n@keyframes mv-check-pulse{0%,100%{opacity:.75;transform:scale(1);}50%{opacity:1;transform:scale(1.12);}}\n.psvr-scope .mv-json-edge{animation:mv-edge-sweep 3.4s linear infinite;}\n@keyframes mv-edge-sweep{0%{stroke-dashoffset:0;}100%{stroke-dashoffset:-192;}}\n.psvr-scope .mv-badge{position:absolute;left:50%;bottom:-4px;transform:translate(-50%,50%);display:inline-flex;align-items:center;gap:8px;font-family:\"JetBrains Mono\",monospace;font-size:11px;font-weight:600;letter-spacing:.02em;color:#eef2fa;background:#0a0d12;border:1px solid rgba(255,255,255,.14);padding:9px 16px;border-radius:100px;white-space:nowrap;box-shadow:0 14px 30px -10px rgba(0,0,0,.55);}\n.psvr-scope .mv-badge i{width:7px;height:7px;border-radius:50%;background:#5bd6a0;box-shadow:0 0 8px 1px rgba(91,214,160,.7);flex:0 0 auto;}\n@media (max-width:560px){.psvr-scope .mv-badge{font-size:10px;padding:7px 12px;white-space:normal;text-align:center;max-width:88%;}}\n@media (prefers-reduced-motion: reduce){.psvr-scope .mv-path,.psvr-scope .mv-doc-anim,.psvr-scope .mv-core-pulse,.psvr-scope .mv-core-facets,.psvr-scope .mv-scanline,.psvr-scope .mv-json-row,.psvr-scope .mv-json-check,.psvr-scope .mv-json-edge,.psvr-scope .mv-spark{animation:none !important;}.psvr-scope .mv-stream{display:none !important;}.psvr-scope .mv-svg animateTransform{display:none;}}\n\n.psvr-scope .grid3{display:grid;grid-template-columns:repeat(3,1fr);gap:22px;}\n@media (max-width:880px){.psvr-scope .grid3{grid-template-columns:1fr 1fr;}}\n@media (max-width:600px){.psvr-scope .grid3{grid-template-columns:1fr;}}\n.psvr-scope .case{background:var(--bg);border:1px solid var(--line);border-radius:var(--radius);padding:26px 24px;box-shadow:var(--shadow);}\n.psvr-scope .case .icon{width:38px;height:38px;border-radius:9px;background:var(--brand-glow);color:var(--brand);display:flex;align-items:center;justify-content:center;font-size:18px;margin-bottom:16px;}\n.psvr-scope .case .icon svg{display:block;width:20px;height:20px;}\n.psvr-scope .case h3{font-size:17px;font-weight:700;margin-bottom:9px;}\n.psvr-scope .case p{font-size:15px;color:var(--ink-soft);}\n\n.psvr-scope .usecase-feature{display:flex;gap:20px;align-items:flex-start;background:var(--bg);border:1px solid rgba(73,135,245,.3);border-left:4px solid var(--brand);border-radius:var(--radius);padding:26px 28px;box-shadow:var(--shadow);margin-bottom:22px;}\n.psvr-scope .usecase-feature .icon{width:44px;height:44px;border-radius:10px;background:var(--brand-glow);color:var(--brand);display:flex;align-items:center;justify-content:center;flex:0 0 auto;}\n.psvr-scope .usecase-feature .icon svg{display:block;width:24px;height:24px;}\n.psvr-scope .usecase-feature .kicker{display:flex;align-items:center;gap:10px;font-family:\"JetBrains Mono\",monospace;font-size:11px;font-weight:700;letter-spacing:.05em;text-transform:uppercase;color:var(--brand);margin-bottom:6px;}\n.psvr-scope .usecase-feature h3{font-size:18px;margin-bottom:8px;}\n.psvr-scope .usecase-feature p{font-size:15.5px;color:var(--ink-soft);max-width:70ch;}\n@media (max-width:600px){.psvr-scope .usecase-feature{flex-direction:column;}}\n\n.psvr-scope .kpiband{display:grid;grid-template-columns:repeat(3,1fr);gap:36px;padding-top:8px;}\n@media (max-width:720px){.psvr-scope .kpiband{grid-template-columns:1fr;gap:28px;}}\n.psvr-scope .kpi .num{font-family:\"JetBrains Mono\",monospace;font-weight:600;font-size:46px;background:linear-gradient(90deg,var(--brand) 0%, var(--teal) 100%);-webkit-background-clip:text;background-clip:text;color:transparent;}\n.psvr-scope .kpi .num.sm{font-size:36px;}\n.psvr-scope .kpi p{font-size:15px;color:var(--ink-soft);margin-top:6px;max-width:30ch;}\n.psvr-scope .kpiband.four{grid-template-columns:repeat(4,1fr);gap:28px;}\n@media (max-width:880px){.psvr-scope .kpiband.four{grid-template-columns:1fr 1fr;}}\n@media (max-width:520px){.psvr-scope .kpiband.four{grid-template-columns:1fr;}}\n\n.psvr-scope .flow{display:grid;grid-template-columns:1fr auto 1fr auto 1fr;gap:0;align-items:center;margin-top:44px;}\n@media (max-width:820px){.psvr-scope .flow{grid-template-columns:1fr;gap:14px;}}\n.psvr-scope .flow .node{background:var(--bg-deep-alt);border:1px solid var(--line-deep);border-radius:9px;padding:20px 18px;text-align:center;}\n.psvr-scope .flow .node.hl{background:linear-gradient(160deg,var(--brand-deep),var(--brand));border-color:var(--brand);}\n.psvr-scope .flow .node .k{font-family:\"JetBrains Mono\",monospace;font-size:10.5px;color:#7dabf9;letter-spacing:.06em;text-transform:uppercase;margin-bottom:8px;}\n.psvr-scope .flow .node.hl .k{color:rgba(255,255,255,.8);}\n.psvr-scope .flow .node .t{font-size:13.5px;font-weight:700;color:var(--invert);}\n.psvr-scope .flow .node .sub{display:block;font-size:11.5px;font-weight:500;color:var(--invert-soft);margin-top:4px;}\n.psvr-scope .flow .node.hl .sub{color:rgba(255,255,255,.75);}\n.psvr-scope .flow .arrow{text-align:center;color:var(--invert-soft);font-size:20px;}\n@media (max-width:820px){.psvr-scope .flow .arrow{transform:rotate(90deg);}}\n.psvr-scope section:not(.deep) .flow .node{background:var(--bg-alt);border-color:var(--line);}\n.psvr-scope section:not(.deep) .flow .node .t{color:var(--ink);}\n.psvr-scope section:not(.deep) .flow .node .sub{color:var(--ink-soft);}\n.psvr-scope section:not(.deep) .flow .node .k{color:var(--brand-deep);}\n.psvr-scope section:not(.deep) .flow .arrow{color:var(--ink-faint);}\n.psvr-scope section:not(.deep) .flow .node.hl{background:linear-gradient(160deg,var(--brand-deep),var(--brand));border-color:var(--brand);}\n.psvr-scope section:not(.deep) .flow .node.hl .t,\n.psvr-scope section:not(.deep) .flow .node.hl .k,\n.psvr-scope section:not(.deep) .flow .node.hl .sub{color:#fff;}\n\n.psvr-scope .grid3.compliance .case{background:var(--bg-deep-alt);border-color:var(--line-deep);}\n.psvr-scope .grid3.compliance .case h3{color:var(--invert);}\n.psvr-scope .grid3.compliance .case p{color:var(--invert-soft);}\n.psvr-scope .grid3.compliance .icon{background:rgba(73,135,245,.18);color:#7dabf9;}\n\n.psvr-scope .tablewrap{overflow-x:auto;border:1px solid var(--line);border-radius:var(--radius);}\n.psvr-scope table.cmp{width:100%;border-collapse:collapse;min-width:680px;background:var(--bg);}\n.psvr-scope table.cmp th, .psvr-scope table.cmp td{padding:18px 20px;text-align:left;border-bottom:1px solid var(--line);vertical-align:top;}\n.psvr-scope table.cmp thead th{font-size:13.5px;font-family:\"JetBrains Mono\",monospace;font-weight:600;color:var(--ink-soft);background:var(--bg-alt);}\n.psvr-scope table.cmp thead th.hl{color:var(--brand);}\n.psvr-scope table.cmp td.feat{max-width:280px;}\n.psvr-scope table.cmp td.feat b{display:block;font-size:14.5px;font-weight:700;color:var(--ink);}\n.psvr-scope table.cmp td{font-size:14px;color:var(--ink-soft);}\n.psvr-scope table.cmp tbody tr:last-child td{border-bottom:none;}\n.psvr-scope .mark{display:inline-flex;align-items:center;gap:6px;font-weight:700;font-size:13.5px;}\n.psvr-scope .mark.yes{color:var(--good);}\n.psvr-scope .mark.no{color:var(--ink-faint);}\n.psvr-scope .mark.plain{color:var(--ink-soft);font-weight:600;}\n.psvr-scope .riskchip-note{display:block;font-size:11.5px;color:var(--ink-faint);margin-top:4px;font-family:\"Barlow\",sans-serif;}\n\n\/* dokumentli-specific additions *\/\n.psvr-scope .tag-soon{display:inline-flex;align-items:center;font-family:\"JetBrains Mono\",monospace;font-size:10px;font-weight:700;letter-spacing:.04em;text-transform:uppercase;color:var(--good);background:rgba(31,157,111,.14);border:1px solid rgba(31,157,111,.3);padding:3px 9px;border-radius:100px;white-space:nowrap;}\n.psvr-scope .deep .tag-soon,.psvr-scope .tag-soon.on-deep{color:#5bd6a0;background:rgba(31,157,111,.16);border-color:rgba(31,157,111,.35);}\n\n.psvr-scope .grid4{display:grid;grid-template-columns:repeat(4,1fr);gap:22px;}\n@media (max-width:1000px){.psvr-scope .grid4{grid-template-columns:1fr 1fr;}}\n@media (max-width:560px){.psvr-scope .grid4{grid-template-columns:1fr;}}\n\n.psvr-scope .barchart-card{background:var(--bg-deep-alt);border:1px solid var(--line-deep);border-radius:var(--radius);padding:28px 26px 20px;}\n.psvr-scope .barchart-card + .barchart-card{margin-top:22px;}\n.psvr-scope .barchart-card .bc-head{display:flex;align-items:flex-start;justify-content:space-between;gap:14px;margin-bottom:24px;flex-wrap:wrap;}\n.psvr-scope .barchart-card .bc-head h3{font-size:15.5px;color:var(--invert);font-weight:700;max-width:56ch;}\n.psvr-scope .barchart-card .bc-flag{font-family:\"JetBrains Mono\",monospace;font-size:10.5px;font-weight:700;letter-spacing:.05em;text-transform:uppercase;color:#7dabf9;background:rgba(73,135,245,.16);border:1px solid rgba(73,135,245,.3);padding:5px 11px;border-radius:100px;white-space:nowrap;flex:0 0 auto;}\n.psvr-scope .barchart-card .bc-axisnote{margin-top:14px;font-size:11.5px;color:var(--invert-soft);}\n.psvr-scope .barchart{display:flex;align-items:flex-end;gap:10px;height:200px;border-bottom:1px solid var(--line-deep);padding-bottom:2px;}\n.psvr-scope .barchart .bar{position:relative;flex:1;display:flex;flex-direction:column;align-items:center;justify-content:flex-end;height:100%;min-width:0;}\n.psvr-scope .barchart .bar .val{font-family:\"JetBrains Mono\",monospace;font-size:12px;color:var(--invert-soft);margin-bottom:7px;white-space:nowrap;}\n.psvr-scope .barchart .bar .fill{width:100%;border-radius:5px 5px 0 0;background:var(--line-deep);}\n.psvr-scope .barchart .bar.best{z-index:1;}\n.psvr-scope .barchart .bar.best::before{content:\"\";position:absolute;top:-38px;left:-8px;right:-8px;bottom:0;background:linear-gradient(180deg,rgba(46,196,182,.14),rgba(46,196,182,0) 75%);border-radius:10px 10px 0 0;z-index:0;}\n.psvr-scope .barchart .bar.best .crown,.psvr-scope .barchart .bar.best .val,.psvr-scope .barchart .bar.best .fill,.psvr-scope .barchart .bar.best .lbl{position:relative;z-index:1;}\n.psvr-scope .barchart .bar.best .crown{display:block;font-family:\"JetBrains Mono\",monospace;font-size:9.5px;font-weight:700;letter-spacing:.04em;text-transform:uppercase;color:var(--teal);margin-bottom:5px;white-space:nowrap;}\n.psvr-scope .barchart .bar.best .fill{background:linear-gradient(180deg,var(--teal),var(--brand));box-shadow:0 12px 26px -8px rgba(46,196,182,.5);}\n.psvr-scope .barchart .bar.best .val{color:var(--invert);font-weight:700;font-size:13.5px;}\n.psvr-scope .barchart .bar .lbl{margin-top:10px;font-size:11px;color:var(--invert-soft);text-align:center;line-height:1.25;}\n.psvr-scope .barchart .bar.best .lbl{color:var(--invert);font-weight:700;}\n@media (max-width:640px){.psvr-scope .barchart{height:190px;}.psvr-scope .barchart .lbl{font-size:9.5px;}}\n\n.psvr-scope .decision-grid{display:grid;grid-template-columns:1fr 1fr;gap:22px;}\n@media (max-width:720px){.psvr-scope .decision-grid{grid-template-columns:1fr;}}\n.psvr-scope .decision{border:1px solid var(--line);border-radius:var(--radius);padding:26px 26px 24px;background:var(--bg);}\n.psvr-scope .decision.yes{border-color:rgba(31,157,111,.35);background:rgba(31,157,111,.05);}\n.psvr-scope .decision .dh{display:flex;align-items:center;gap:12px;margin-bottom:16px;}\n.psvr-scope .decision .dh .icon{width:36px;height:36px;border-radius:9px;display:flex;align-items:center;justify-content:center;flex:0 0 auto;}\n.psvr-scope .decision.yes .dh .icon{background:rgba(31,157,111,.14);color:var(--good);}\n.psvr-scope .decision.no .dh .icon{background:var(--bg-alt);color:var(--ink-soft);}\n.psvr-scope .decision .dh .icon svg{width:19px;height:19px;}\n.psvr-scope .decision .dh h3{font-size:16.5px;}\n.psvr-scope .decision ul{margin:14px 0 0;padding-left:20px;display:flex;flex-direction:column;gap:9px;}\n.psvr-scope .decision li{font-size:14.5px;color:var(--ink-soft);}\n.psvr-scope .decision.plain{border-color:var(--line);}\n.psvr-scope .decision .kicker{display:block;font-family:\"JetBrains Mono\",monospace;font-size:11px;font-weight:700;letter-spacing:.05em;text-transform:uppercase;color:var(--brand);margin-bottom:8px;}\n.psvr-scope .decision ul.plain{padding-left:0;list-style:none !important;}\n.psvr-scope .decision ul.plain li{padding-left:18px;position:relative;}\n.psvr-scope .decision ul.plain li::before{content:\"\";position:absolute;left:0;top:7px;width:5px;height:5px;border-radius:50%;background:var(--brand);}\n.psvr-scope .decision ul.plain li b{color:var(--ink);font-weight:600;}\n\n.psvr-scope .note{display:flex;gap:16px;align-items:flex-start;border-radius:var(--radius);padding:24px 26px;}\n.psvr-scope .note.brand{background:linear-gradient(120deg,var(--brand-deep),var(--brand) 140%);color:#fff;}\n.psvr-scope .note.brand h3{color:#fff;}\n.psvr-scope .note.brand p{color:rgba(255,255,255,.9);}\n.psvr-scope .note.plain{background:var(--bg-alt);border:1px solid var(--line);}\n.psvr-scope .note.deepbox{background:var(--bg-deep-alt);border:1px solid var(--line-deep);}\n.psvr-scope .note.deepbox h3{color:var(--invert);}\n.psvr-scope .note.deepbox p{color:var(--invert-soft);}\n.psvr-scope .note .icon{width:38px;height:38px;border-radius:10px;background:rgba(255,255,255,.18);display:flex;align-items:center;justify-content:center;flex:0 0 auto;color:#fff;}\n.psvr-scope .note.plain .icon{background:var(--brand-glow);color:var(--brand);}\n.psvr-scope .note.deepbox .icon{background:rgba(73,135,245,.18);color:#7dabf9;}\n.psvr-scope .note .icon svg{width:20px;height:20px;}\n.psvr-scope .note h3{font-size:16px;margin-bottom:6px;}\n.psvr-scope .note p{font-size:14.5px;line-height:1.55;}\n.psvr-scope .note .kicker{display:block;font-family:\"JetBrains Mono\",monospace;font-size:11px;font-weight:700;letter-spacing:.05em;text-transform:uppercase;margin-bottom:5px;opacity:.85;}\n\n.psvr-scope .miniflow{display:flex;align-items:center;gap:7px;margin-bottom:20px;flex-wrap:wrap;}\n.psvr-scope .miniflow .mf-node{width:34px;height:34px;border-radius:8px;background:var(--brand-glow);color:var(--brand);display:flex;align-items:center;justify-content:center;flex:0 0 auto;}\n.psvr-scope .miniflow .mf-node svg{width:16px;height:16px;}\n.psvr-scope .miniflow .mf-arrow{color:var(--ink-faint);font-size:14px;}\n\n.psvr-scope .deploy-card{background:var(--bg-deep-alt);border:1px solid var(--line-deep);border-radius:var(--radius);padding:26px 24px;display:flex;flex-direction:column;}\n.psvr-scope .deploy-card.hl{border-color:var(--brand);box-shadow:0 0 0 1px var(--brand);}\n.psvr-scope .deploy-card .icon{width:40px;height:40px;border-radius:10px;background:rgba(73,135,245,.18);color:#7dabf9;display:flex;align-items:center;justify-content:center;margin-bottom:14px;}\n.psvr-scope .deploy-card .icon svg{width:21px;height:21px;}\n.psvr-scope .deploy-card h3{font-size:16.5px;margin-bottom:12px;color:var(--invert);}\n.psvr-scope .deploy-card .dtag{display:inline-block;font-family:\"JetBrains Mono\",monospace;font-size:10px;font-weight:700;letter-spacing:.05em;text-transform:uppercase;color:#9db8f7;background:rgba(73,135,245,.16);padding:4px 10px;border-radius:100px;margin-bottom:14px;align-self:flex-start;}\n.psvr-scope .deploy-card ul{margin:0;padding:0;list-style:none !important;display:flex;flex-direction:column;gap:10px;}\n.psvr-scope .deploy-card li{font-size:13.5px;color:var(--invert-soft);padding-left:16px;position:relative;}\n.psvr-scope .deploy-card li::before{content:\"\";position:absolute;left:0;top:7px;width:5px;height:5px;border-radius:50%;background:#7dabf9;}\n.psvr-scope .deploy-card li b{color:var(--invert);font-weight:600;}\n\n.psvr-scope .genspec{display:grid;grid-template-columns:1fr 1fr;gap:20px;margin-top:26px;}\n@media (max-width:640px){.psvr-scope .genspec{grid-template-columns:1fr;}}\n.psvr-scope .genspec > div{border:1px solid var(--line);border-radius:var(--radius);padding:22px 24px;}\n.psvr-scope .genspec > div.spec{border-color:rgba(73,135,245,.35);background:var(--brand-glow);}\n.psvr-scope .genspec h4{font-size:15px;margin-bottom:8px;}\n.psvr-scope .genspec p{font-size:14px;color:var(--ink-soft);}\n\n.psvr-scope .costcard{background:var(--bg);border:1px solid var(--line);border-radius:var(--radius);padding:28px 26px 20px;box-shadow:var(--shadow);}\n.psvr-scope .costcard .cc-head h3{font-size:15.5px;margin-bottom:6px;}\n.psvr-scope .costcard .cc-head p{font-size:13px;color:var(--ink-faint);max-width:64ch;margin-bottom:18px;}\n.psvr-scope .costlegend{display:flex;gap:22px;margin-bottom:18px;font-size:12.5px;color:var(--ink-soft);flex-wrap:wrap;}\n.psvr-scope .costlegend span{display:flex;align-items:center;gap:7px;}\n.psvr-scope .costlegend i{width:11px;height:11px;border-radius:3px;display:inline-block;}\n.psvr-scope .costlegend i.ppc{background:#cdd4e2;}\n.psvr-scope .costlegend i.sub{background:linear-gradient(180deg,var(--teal),var(--brand));}\n.psvr-scope .costchart{display:flex;align-items:flex-end;gap:8px;height:150px;border-bottom:1px solid var(--line);padding-bottom:2px;}\n.psvr-scope .costchart .cgroup{flex:1;display:flex;align-items:flex-end;gap:4px;height:100%;}\n.psvr-scope .costchart .cbar{flex:1;border-radius:4px 4px 0 0;}\n.psvr-scope .costchart .cbar.ppc{background:#cdd4e2;}\n.psvr-scope .costchart .cbar.sub{background:linear-gradient(180deg,var(--teal),var(--brand));}\n.psvr-scope .costaxis{display:flex;gap:8px;margin-top:10px;}\n.psvr-scope .costaxis span{flex:1;text-align:center;font-family:\"JetBrains Mono\",monospace;font-size:10.5px;color:var(--ink-faint);}\n.psvr-scope .cost-flag{margin-left:auto;font-family:\"JetBrains Mono\",monospace;font-size:10.5px;font-weight:700;letter-spacing:.04em;text-transform:uppercase;color:var(--ink-faint);}\n\n.psvr-scope .faq{max-width:760px;}\n.psvr-scope .faqitem{border-bottom:1px solid var(--line);}\n.psvr-scope .faqitem:first-child{border-top:1px solid var(--line);}\n.psvr-scope .faqitem button.q{width:100%;display:flex;align-items:center;justify-content:space-between;gap:16px;background:none;border:none;text-align:left;padding:20px 4px;cursor:pointer;font-family:\"Barlow\",sans-serif;font-size:16px;font-weight:700;color:var(--ink);}\n.psvr-scope .faqitem .q .plus{font-family:\"JetBrains Mono\",monospace;color:var(--brand);font-size:18px;flex:0 0 auto;transition:transform .18s ease;}\n.psvr-scope .faqitem.open .q .plus{transform:rotate(45deg);}\n.psvr-scope .faqitem .a{max-height:0;overflow:hidden;transition:max-height .22s ease;}\n.psvr-scope .faqitem.open .a{max-height:320px;}\n.psvr-scope .faqitem .a p{padding:0 4px 20px;font-size:15.5px;color:var(--ink-soft);max-width:68ch;}\n\n.psvr-scope .finalcta{background:linear-gradient(120deg,var(--brand-deep),var(--brand) 60%, var(--teal) 130%);border-radius:20px;padding:56px 48px;text-align:center;color:#fff;}\n@media (max-width:720px){.psvr-scope .finalcta{padding:40px 24px;}}\n.psvr-scope .finalcta h2{color:#fff;font-size:clamp(24px,3vw,32px);margin-bottom:14px;}\n.psvr-scope .finalcta p{color:rgba(255,255,255,.85);max-width:52ch;margin:0 auto 28px;font-size:16px;}\n.psvr-scope .finalcta .ctas{display:flex;gap:14px;justify-content:center;flex-wrap:wrap;}\n.psvr-scope .finalcta .btn{background:#fff;color:var(--brand-deep);border-color:#fff;box-shadow:0 10px 24px -10px rgba(0,0,0,.35);}\n.psvr-scope .finalcta .btn:hover{background:#f2f6ff;color:var(--brand-deep);}\n.psvr-scope .finalcta .btn.ghost{background:transparent;color:#fff;border-color:rgba(255,255,255,.4);box-shadow:none;}\n.psvr-scope .finalcta .btn.ghost:hover{background:rgba(255,255,255,.12);color:#fff;}\n.psvr-scope .finalcta .cite{background:rgba(255,255,255,.14);border-color:rgba(255,255,255,.25);color:#fff;margin-top:22px;}\n.psvr-scope .finalcta .cite b{color:#fff;}\n\n.psvr-scope .linkcards{display:grid;grid-template-columns:repeat(3,1fr);gap:20px;margin-top:40px;}\n@media (max-width:820px){.psvr-scope .linkcards{grid-template-columns:1fr;}}\n.psvr-scope .linkcard{display:flex;gap:16px;align-items:flex-start;background:var(--bg);border:1px solid var(--line);border-radius:var(--radius);padding:22px;box-shadow:var(--shadow);transition:border-color .15s ease, transform .15s ease;}\n.psvr-scope a.linkcard:hover{border-color:var(--brand);transform:translateY(-2px);}\n.psvr-scope .linkcard .icon{width:36px;height:36px;border-radius:9px;background:var(--brand-glow);color:var(--brand);display:flex;align-items:center;justify-content:center;flex:0 0 auto;}\n.psvr-scope .linkcard .icon svg{width:19px;height:19px;}\n.psvr-scope .linkcard h4{font-size:15px;margin-bottom:5px;}\n.psvr-scope .linkcard p{font-size:13.5px;color:var(--ink-soft);}\n.psvr-scope .linkcard .go{color:var(--brand);font-weight:700;font-size:13px;margin-top:6px;display:inline-block;}\n\n.psvr-scope .compliance-badges{margin-top:40px;padding-top:32px;border-top:1px solid var(--line-deep);}\n.psvr-scope .compliance-badges .badges-label{display:block;font-family:\"JetBrains Mono\",monospace;font-size:11.5px;font-weight:600;letter-spacing:.06em;text-transform:uppercase;color:var(--invert-soft);margin-bottom:16px;}\n.psvr-scope .compliance-badges .badges-row{display:flex;gap:16px;flex-wrap:wrap;align-items:center;}\n.psvr-scope .compliance-badges .badge-logo{width:64px;height:64px;background:#fff;border-radius:12px;padding:8px;display:flex;align-items:center;justify-content:center;box-shadow:0 6px 16px -8px rgba(0,0,0,.35);flex:0 0 auto;}\n.psvr-scope .compliance-badges .badge-logo img{width:100%;height:100%;object-fit:contain;}\n\n.psvr-scope::selection{background:var(--brand-glow);color:var(--brand-deep);}\n@media (prefers-reduced-motion: reduce){.psvr-scope *{transition:none !important;}}\n<\/style>\n\n<div class=\"psvr-scope\">\n\n<section class=\"hero\">\n  <div class=\"wrap\">\n    <div class=\"hero-grid\">\n      <div>\n        <span class=\"eyebrow\">Dokumentli\u00ae \u00b7 Sovereign Document AI<\/span>\n        <h1>Dokumentli\u00ae \u2013 Das KI-Modell, das LLMs bei Dokumenten schl\u00e4gt.<\/h1>\n        <p class=\"lede\">Dokumentli\u00ae liest, klassifiziert und strukturiert Gesch\u00e4ftsdokumente pr\u00e4ziser und schneller als general-purpose Frontier-Modelle \u2013 und l\u00e4uft wahlweise vollst\u00e4ndig air-gapped auf einer einzelnen Consumer-GPU mit 24 GB VRAM.<\/p>\n        <div class=\"ctas\">\n          <a class=\"btn\" href=\"https:\/\/playground.parashift.io\/\" target=\"_blank\" rel=\"noopener\">Modell live testen<\/a>\n          <a class=\"btn ghost\" href=\"#benchmarks\">Benchmark-Ergebnisse ansehen \u2193<\/a>\n        <\/div>\n        <div class=\"microproof\">\n          <span><i class=\"dot\"><\/i> Schneller<\/span>\n          <span><i class=\"dot\"><\/i> Genauer<\/span>\n          <span><i class=\"dot\"><\/i> G\u00fcnstiger<\/span>\n          <span><i class=\"dot\"><\/i> L\u00e4uft auf 1 GPU (24 GB)<\/span>\n        <\/div>\n      <\/div>\n      <div class=\"ai-aura\">\n        <span class=\"aura-vignette\"><\/span>\n        <span class=\"aura-stars s2\"><\/span>\n        <span class=\"aura-stars\"><\/span>\n        <span class=\"aura-nebula2\"><\/span>\n        <span class=\"aura-glow\"><\/span>\n        <span class=\"aura-grid\"><\/span>\n        <span class=\"aura-radar\"><\/span>\n        <span class=\"aura-corona\"><\/span>\n        <span class=\"pulse-ring r1\"><\/span>\n        <span class=\"pulse-ring r2\"><\/span>\n        <span class=\"pulse-ring r3\"><\/span>\n        <div class=\"model-visual\">\n          <svg class=\"mv-svg\" viewBox=\"0 0 640 560\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" role=\"img\" aria-label=\"Animierte Visualisierung: Dokumentli scannt fortlaufend eingehende Dokumente, verarbeitet sie in einem rotierenden KI-Kern und gibt strukturierte, validierte Daten aus\">\n            <defs>\n              <radialGradient id=\"mvCoreGlow\" cx=\"50%\" cy=\"50%\" r=\"50%\">\n                <stop offset=\"0%\" stop-color=\"#e4f4ff\" stop-opacity=\".98\"\/>\n                <stop offset=\"35%\" stop-color=\"#5ee0ff\" stop-opacity=\".5\"\/>\n                <stop offset=\"68%\" stop-color=\"#8a6bff\" stop-opacity=\".26\"\/>\n                <stop offset=\"100%\" stop-color=\"#8a6bff\" stop-opacity=\"0\"\/>\n              <\/radialGradient>\n              <linearGradient id=\"mvCoreFill\" x1=\"0\" y1=\"0\" x2=\"1\" y2=\"1\">\n                <stop offset=\"0%\" stop-color=\"#c9f2ff\"\/>\n                <stop offset=\"42%\" stop-color=\"#5ee0ff\"\/>\n                <stop offset=\"72%\" stop-color=\"#4987f5\"\/>\n                <stop offset=\"100%\" stop-color=\"#9d4fe0\"\/>\n              <\/linearGradient>\n              <linearGradient id=\"flowLine\" x1=\"0\" y1=\"0\" x2=\"1\" y2=\"0\">\n                <stop offset=\"0%\" stop-color=\"#4987f5\" stop-opacity=\"0\"\/>\n                <stop offset=\"42%\" stop-color=\"#5ee0ff\" stop-opacity=\".95\"\/>\n                <stop offset=\"58%\" stop-color=\"#c98bff\" stop-opacity=\".95\"\/>\n                <stop offset=\"100%\" stop-color=\"#6d4fe0\" stop-opacity=\"0\"\/>\n              <\/linearGradient>\n              <linearGradient id=\"scanGrad\" x1=\"0\" y1=\"0\" x2=\"0\" y2=\"1\">\n                <stop offset=\"0%\" stop-color=\"#5ee0ff\" stop-opacity=\"0\"\/>\n                <stop offset=\"50%\" stop-color=\"#bfe6ff\" stop-opacity=\".95\"\/>\n                <stop offset=\"100%\" stop-color=\"#5ee0ff\" stop-opacity=\"0\"\/>\n              <\/linearGradient>\n              <clipPath id=\"docClip\"><rect width=\"80\" height=\"96\" rx=\"7\"\/><\/clipPath>\n              <filter id=\"mvGlow\" x=\"-80%\" y=\"-80%\" width=\"260%\" height=\"260%\">\n                <feGaussianBlur stdDeviation=\"3.2\" result=\"b\"\/>\n                <feMerge><feMergeNode in=\"b\"\/><feMergeNode in=\"SourceGraphic\"\/><\/feMerge>\n              <\/filter>\n            <\/defs>\n\n            <path class=\"mv-path-glow\" d=\"M118,132 C230,140 246,224 318,272\"\/>\n            <path class=\"mv-path-glow\" d=\"M100,296 C220,300 252,290 318,278\"\/>\n            <path class=\"mv-path-glow\" d=\"M120,458 C236,452 258,344 318,286\"\/>\n            <path class=\"mv-path-glow\" d=\"M322,278 C398,232 462,208 538,150\"\/>\n            <path class=\"mv-path\" d=\"M118,132 C230,140 246,224 318,272\"\/>\n            <path class=\"mv-path\" d=\"M100,296 C220,300 252,290 318,278\"\/>\n            <path class=\"mv-path\" d=\"M120,458 C236,452 258,344 318,286\"\/>\n            <path class=\"mv-path out\" d=\"M322,278 C398,232 462,208 538,150\"\/>\n\n            <circle class=\"mv-core-pulse\" cx=\"320\" cy=\"280\" r=\"150\" fill=\"url(#mvCoreGlow)\"\/>\n\n            <!-- armillary orbit rings: three tilted ellipses spinning on independent axes, atom\/gyroscope style -->\n            <g transform=\"rotate(-28 320 280)\">\n              <g>\n                <animateTransform attributeName=\"transform\" type=\"rotate\" from=\"0 320 280\" to=\"360 320 280\" dur=\"7s\" repeatCount=\"indefinite\"\/>\n                <ellipse cx=\"320\" cy=\"280\" rx=\"118\" ry=\"42\" fill=\"none\" stroke=\"rgba(157,199,255,.4)\" stroke-width=\"1\"\/>\n                <circle class=\"mv-node\" cx=\"438\" cy=\"280\" r=\"3\" fill=\"#5ee0ff\"\/>\n              <\/g>\n            <\/g>\n            <g transform=\"rotate(35 320 280)\">\n              <g>\n                <animateTransform attributeName=\"transform\" type=\"rotate\" from=\"360 320 280\" to=\"0 320 280\" dur=\"9.5s\" repeatCount=\"indefinite\"\/>\n                <ellipse cx=\"320\" cy=\"280\" rx=\"126\" ry=\"48\" fill=\"none\" stroke=\"rgba(201,139,255,.42)\" stroke-width=\"1\"\/>\n                <circle class=\"mv-node\" cx=\"446\" cy=\"280\" r=\"2.6\" fill=\"#c98bff\"\/>\n              <\/g>\n            <\/g>\n            <g transform=\"rotate(84 320 280)\">\n              <g>\n                <animateTransform attributeName=\"transform\" type=\"rotate\" from=\"0 320 280\" to=\"360 320 280\" dur=\"12s\" repeatCount=\"indefinite\"\/>\n                <ellipse cx=\"320\" cy=\"280\" rx=\"108\" ry=\"38\" fill=\"none\" stroke=\"rgba(157,199,255,.5)\" stroke-width=\"1.1\"\/>\n                <circle class=\"mv-node\" cx=\"428\" cy=\"280\" r=\"2.3\" fill=\"#9dc7ff\"\/>\n              <\/g>\n            <\/g>\n\n            <circle cx=\"320\" cy=\"280\" r=\"56\" fill=\"url(#mvCoreFill)\"\/>\n            <circle cx=\"320\" cy=\"280\" r=\"56\" fill=\"none\" stroke=\"rgba(255,255,255,.4)\" stroke-width=\"1\"\/>\n\n            <g class=\"mv-core-facets\" filter=\"url(#mvGlow)\">\n              <polygon points=\"320,236 358,261 358,299 320,324 282,299 282,261\" fill=\"none\" stroke=\"rgba(255,255,255,.6)\" stroke-width=\"1\"\/>\n              <polygon points=\"320,236 358,261 320,280\" fill=\"rgba(228,244,255,.3)\"\/>\n              <polygon points=\"358,261 358,299 320,280\" fill=\"rgba(228,244,255,.13)\"\/>\n              <polygon points=\"358,299 320,324 320,280\" fill=\"rgba(228,244,255,.34)\"\/>\n              <polygon points=\"320,324 282,299 320,280\" fill=\"rgba(228,244,255,.11)\"\/>\n              <polygon points=\"282,299 282,261 320,280\" fill=\"rgba(228,244,255,.26)\"\/>\n              <polygon points=\"282,261 320,236 320,280\" fill=\"rgba(228,244,255,.18)\"\/>\n              <circle cx=\"320\" cy=\"280\" r=\"5\" fill=\"#fff\"\/>\n            <\/g>\n            <g class=\"mv-core-facets rev\" opacity=\".65\">\n              <polygon points=\"320,248 348,264 348,296 320,312 292,296 292,264\" fill=\"none\" stroke=\"rgba(94,224,255,.6)\" stroke-width=\".8\"\/>\n            <\/g>\n\n            <g fill=\"#fff\">\n              <circle class=\"mv-spark\" cx=\"210\" cy=\"160\" r=\"1.4\" style=\"animation-delay:.2s\"\/>\n              <circle class=\"mv-spark\" cx=\"470\" cy=\"200\" r=\"1.6\" style=\"animation-delay:1.1s\"\/>\n              <circle class=\"mv-spark\" cx=\"150\" cy=\"380\" r=\"1.3\" style=\"animation-delay:1.8s\"\/>\n              <circle class=\"mv-spark\" cx=\"500\" cy=\"410\" r=\"1.5\" style=\"animation-delay:.7s\"\/>\n              <circle class=\"mv-spark\" cx=\"230\" cy=\"480\" r=\"1.2\" style=\"animation-delay:2.3s\"\/>\n              <circle class=\"mv-spark\" cx=\"420\" cy=\"120\" r=\"1.3\" style=\"animation-delay:1.5s\"\/>\n            <\/g>\n\n            <g class=\"mv-stream-doc\">\n              <g class=\"mv-stream p-top t0\">\n                <rect x=\"-11\" y=\"-13.5\" width=\"22\" height=\"27\" rx=\"4\" fill=\"#fff\"\/>\n                <rect x=\"-7\" y=\"-7.5\" width=\"10\" height=\"2.6\" rx=\"1.3\" fill=\"#9fb3d4\"\/>\n                <rect x=\"-7\" y=\"-2.1\" width=\"14\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n                <rect x=\"-7\" y=\"3.3\" width=\"8\" height=\"2.6\" rx=\"1.3\" fill=\"#4987f5\"\/>\n                <rect x=\"-7\" y=\"8.7\" width=\"12\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n              <\/g>\n              <g class=\"mv-stream p-top t5\">\n                <rect x=\"-11\" y=\"-13.5\" width=\"22\" height=\"27\" rx=\"4\" fill=\"#fff\"\/>\n                <rect x=\"-7\" y=\"-7.5\" width=\"10\" height=\"2.6\" rx=\"1.3\" fill=\"#9fb3d4\"\/>\n                <rect x=\"-7\" y=\"-2.1\" width=\"14\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n                <rect x=\"-7\" y=\"3.3\" width=\"8\" height=\"2.6\" rx=\"1.3\" fill=\"#4987f5\"\/>\n                <rect x=\"-7\" y=\"8.7\" width=\"12\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n              <\/g>\n              <g class=\"mv-stream p-top t9\">\n                <rect x=\"-11\" y=\"-13.5\" width=\"22\" height=\"27\" rx=\"4\" fill=\"#fff\"\/>\n                <rect x=\"-7\" y=\"-7.5\" width=\"10\" height=\"2.6\" rx=\"1.3\" fill=\"#9fb3d4\"\/>\n                <rect x=\"-7\" y=\"-2.1\" width=\"14\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n                <rect x=\"-7\" y=\"3.3\" width=\"8\" height=\"2.6\" rx=\"1.3\" fill=\"#4987f5\"\/>\n                <rect x=\"-7\" y=\"8.7\" width=\"12\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n              <\/g>\n\n              <g class=\"mv-stream p-mid t1\">\n                <rect x=\"-10\" y=\"-12.25\" width=\"20\" height=\"24.5\" rx=\"4\" fill=\"#fff\"\/>\n                <rect x=\"-6.5\" y=\"-6.75\" width=\"9\" height=\"2.4\" rx=\"1.2\" fill=\"#9fb3d4\"\/>\n                <rect x=\"-6.5\" y=\"-1.85\" width=\"13\" height=\"2.4\" rx=\"1.2\" fill=\"#e2e8f2\"\/>\n                <rect x=\"-6.5\" y=\"3.05\" width=\"7\" height=\"2.4\" rx=\"1.2\" fill=\"#4987f5\"\/>\n                <rect x=\"-6.5\" y=\"7.95\" width=\"11\" height=\"2.4\" rx=\"1.2\" fill=\"#e2e8f2\"\/>\n              <\/g>\n              <g class=\"mv-stream p-mid t7\">\n                <rect x=\"-10\" y=\"-12.25\" width=\"20\" height=\"24.5\" rx=\"4\" fill=\"#fff\"\/>\n                <rect x=\"-6.5\" y=\"-6.75\" width=\"9\" height=\"2.4\" rx=\"1.2\" fill=\"#9fb3d4\"\/>\n                <rect x=\"-6.5\" y=\"-1.85\" width=\"13\" height=\"2.4\" rx=\"1.2\" fill=\"#e2e8f2\"\/>\n                <rect x=\"-6.5\" y=\"3.05\" width=\"7\" height=\"2.4\" rx=\"1.2\" fill=\"#4987f5\"\/>\n                <rect x=\"-6.5\" y=\"7.95\" width=\"11\" height=\"2.4\" rx=\"1.2\" fill=\"#e2e8f2\"\/>\n              <\/g>\n              <g class=\"mv-stream p-mid t10\">\n                <rect x=\"-10\" y=\"-12.25\" width=\"20\" height=\"24.5\" rx=\"4\" fill=\"#fff\"\/>\n                <rect x=\"-6.5\" y=\"-6.75\" width=\"9\" height=\"2.4\" rx=\"1.2\" fill=\"#9fb3d4\"\/>\n                <rect x=\"-6.5\" y=\"-1.85\" width=\"13\" height=\"2.4\" rx=\"1.2\" fill=\"#e2e8f2\"\/>\n                <rect x=\"-6.5\" y=\"3.05\" width=\"7\" height=\"2.4\" rx=\"1.2\" fill=\"#4987f5\"\/>\n                <rect x=\"-6.5\" y=\"7.95\" width=\"11\" height=\"2.4\" rx=\"1.2\" fill=\"#e2e8f2\"\/>\n              <\/g>\n\n              <g class=\"mv-stream p-bot t0\">\n                <rect x=\"-11\" y=\"-13.5\" width=\"22\" height=\"27\" rx=\"4\" fill=\"#fff\"\/>\n                <rect x=\"-7\" y=\"-7.5\" width=\"10\" height=\"2.6\" rx=\"1.3\" fill=\"#9fb3d4\"\/>\n                <rect x=\"-7\" y=\"-2.1\" width=\"14\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n                <rect x=\"-7\" y=\"3.3\" width=\"8\" height=\"2.6\" rx=\"1.3\" fill=\"#4987f5\"\/>\n                <rect x=\"-7\" y=\"8.7\" width=\"12\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n              <\/g>\n              <g class=\"mv-stream p-bot t6\">\n                <rect x=\"-11\" y=\"-13.5\" width=\"22\" height=\"27\" rx=\"4\" fill=\"#fff\"\/>\n                <rect x=\"-7\" y=\"-7.5\" width=\"10\" height=\"2.6\" rx=\"1.3\" fill=\"#9fb3d4\"\/>\n                <rect x=\"-7\" y=\"-2.1\" width=\"14\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n                <rect x=\"-7\" y=\"3.3\" width=\"8\" height=\"2.6\" rx=\"1.3\" fill=\"#4987f5\"\/>\n                <rect x=\"-7\" y=\"8.7\" width=\"12\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n              <\/g>\n              <g class=\"mv-stream p-bot t11\">\n                <rect x=\"-11\" y=\"-13.5\" width=\"22\" height=\"27\" rx=\"4\" fill=\"#fff\"\/>\n                <rect x=\"-7\" y=\"-7.5\" width=\"10\" height=\"2.6\" rx=\"1.3\" fill=\"#9fb3d4\"\/>\n                <rect x=\"-7\" y=\"-2.1\" width=\"14\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n                <rect x=\"-7\" y=\"3.3\" width=\"8\" height=\"2.6\" rx=\"1.3\" fill=\"#4987f5\"\/>\n                <rect x=\"-7\" y=\"8.7\" width=\"12\" height=\"2.6\" rx=\"1.3\" fill=\"#e2e8f2\"\/>\n              <\/g>\n            <\/g>\n\n            <g fill=\"#bcd6ff\">\n              <circle class=\"mv-stream dot p-top t2\" r=\"2.1\"\/>\n              <circle class=\"mv-stream dot p-mid t4\" r=\"2.1\"\/>\n              <circle class=\"mv-stream dot p-bot t8\" r=\"2.1\"\/>\n            <\/g>\n\n            <g fill=\"#5bd6a0\">\n              <circle class=\"mv-stream dot p-out t0\" r=\"2.3\"\/>\n              <circle class=\"mv-stream dot p-out t3\" r=\"2.3\"\/>\n              <circle class=\"mv-stream dot p-out t7\" r=\"2.3\"\/>\n            <\/g>\n\n            <g class=\"mv-doc-anim d1\" transform=\"translate(52,92)\">\n              <rect width=\"80\" height=\"96\" rx=\"7\" fill=\"#fff\"\/>\n              <rect x=\"14\" y=\"18\" width=\"40\" height=\"5\" rx=\"2.5\" fill=\"#9fb3d4\"\/>\n              <rect x=\"14\" y=\"30\" width=\"52\" height=\"5\" rx=\"2.5\" fill=\"#e2e8f2\"\/>\n              <rect x=\"14\" y=\"42\" width=\"52\" height=\"5\" rx=\"2.5\" fill=\"#e2e8f2\"\/>\n              <rect x=\"14\" y=\"54\" width=\"30\" height=\"5\" rx=\"2.5\" fill=\"#4987f5\"\/>\n              <rect x=\"14\" y=\"66\" width=\"44\" height=\"5\" rx=\"2.5\" fill=\"#e2e8f2\"\/>\n              <g clip-path=\"url(#docClip)\"><rect class=\"mv-scanline\" x=\"0\" y=\"0\" width=\"80\" height=\"16\" fill=\"url(#scanGrad)\"\/><\/g>\n              <rect width=\"80\" height=\"96\" rx=\"7\" fill=\"none\" stroke=\"rgba(94,224,255,.0)\"\/>\n            <\/g>\n            <g class=\"mv-doc-anim d2\" transform=\"translate(34,258)\">\n              <rect width=\"80\" height=\"96\" rx=\"7\" fill=\"#fff\"\/>\n              <rect x=\"14\" y=\"18\" width=\"40\" height=\"5\" rx=\"2.5\" fill=\"#9fb3d4\"\/>\n              <rect x=\"14\" y=\"30\" width=\"52\" height=\"5\" rx=\"2.5\" fill=\"#e2e8f2\"\/>\n              <rect x=\"14\" y=\"42\" width=\"30\" height=\"5\" rx=\"2.5\" fill=\"#4987f5\"\/>\n              <rect x=\"14\" y=\"54\" width=\"52\" height=\"5\" rx=\"2.5\" fill=\"#e2e8f2\"\/>\n              <rect x=\"14\" y=\"66\" width=\"38\" height=\"5\" rx=\"2.5\" fill=\"#e2e8f2\"\/>\n              <g clip-path=\"url(#docClip)\"><rect class=\"mv-scanline\" x=\"0\" y=\"0\" width=\"80\" height=\"16\" fill=\"url(#scanGrad)\"\/><\/g>\n            <\/g>\n            <g class=\"mv-doc-anim d3\" transform=\"translate(58,414)\">\n              <rect width=\"80\" height=\"96\" rx=\"7\" fill=\"#fff\"\/>\n              <rect x=\"14\" y=\"18\" width=\"40\" height=\"5\" rx=\"2.5\" fill=\"#9fb3d4\"\/>\n              <rect x=\"14\" y=\"30\" width=\"52\" height=\"5\" rx=\"2.5\" fill=\"#e2e8f2\"\/>\n              <rect x=\"14\" y=\"42\" width=\"52\" height=\"5\" rx=\"2.5\" fill=\"#e2e8f2\"\/>\n              <rect x=\"14\" y=\"54\" width=\"44\" height=\"5\" rx=\"2.5\" fill=\"#e2e8f2\"\/>\n              <rect x=\"14\" y=\"66\" width=\"26\" height=\"5\" rx=\"2.5\" fill=\"#4987f5\"\/>\n              <g clip-path=\"url(#docClip)\"><rect class=\"mv-scanline\" x=\"0\" y=\"0\" width=\"80\" height=\"16\" fill=\"url(#scanGrad)\"\/><\/g>\n            <\/g>\n\n            <g transform=\"translate(534,104)\">\n              <rect width=\"96\" height=\"96\" rx=\"8\" fill=\"#0a0d12\" stroke=\"rgba(255,255,255,.14)\"\/>\n              <rect class=\"mv-json-edge\" x=\"1\" y=\"1\" width=\"94\" height=\"94\" rx=\"7\" fill=\"none\" stroke=\"url(#flowLine)\" stroke-width=\"1.4\" stroke-dasharray=\"26 166\"\/>\n              <text x=\"14\" y=\"26\" font-family=\"JetBrains Mono, monospace\" font-size=\"11\" fill=\"#5bd6a0\">{ }<\/text>\n              <g class=\"mv-json-row r1\"><rect x=\"14\" y=\"36\" width=\"26\" height=\"4\" rx=\"2\" fill=\"rgba(255,255,255,.35)\"\/><rect x=\"44\" y=\"36\" width=\"34\" height=\"4\" rx=\"2\" fill=\"#9dc7ff\"\/><\/g>\n              <g class=\"mv-json-row r2\"><rect x=\"14\" y=\"46\" width=\"22\" height=\"4\" rx=\"2\" fill=\"rgba(255,255,255,.35)\"\/><rect x=\"40\" y=\"46\" width=\"40\" height=\"4\" rx=\"2\" fill=\"#9dc7ff\"\/><\/g>\n              <g class=\"mv-json-row r3\"><rect x=\"14\" y=\"56\" width=\"30\" height=\"4\" rx=\"2\" fill=\"rgba(255,255,255,.35)\"\/><rect x=\"48\" y=\"56\" width=\"32\" height=\"4\" rx=\"2\" fill=\"#9dc7ff\"\/><\/g>\n              <g class=\"mv-json-check\">\n                <circle cx=\"76\" cy=\"78\" r=\"9.5\" fill=\"rgba(31,157,111,.2)\" stroke=\"#5bd6a0\" stroke-width=\"1.2\"\/>\n                <path d=\"M72,78.5 L75,81.5 L81,74.5\" stroke=\"#5bd6a0\" stroke-width=\"1.6\" fill=\"none\" stroke-linecap=\"round\" stroke-linejoin=\"round\"\/>\n              <\/g>\n            <\/g>\n          <\/svg>\n          <span class=\"mv-badge\"><i><\/i>Air-Gapped-f\u00e4hig \u00b7 1 GPU \u00b7 24 GB VRAM<\/span>\n        <\/div>\n      <\/div>\n    <\/div>\n  <\/div>\n<\/section>\n\n<section>\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\">Auf einen Blick<\/span>\n      <h2>L\u00e4uft auf 1 GPU mit 24 GB. Schl\u00e4gt sieben Modelle bei Dokumenten.<\/h2>\n      <p>Gebaut f\u00fcr europ\u00e4ische Unternehmen und regulierte Branchen mit hohem, wiederkehrendem Dokumentenvolumen, von Finanzdienstleistungen und Versicherungen bis Bau, Handel und Logistik.<\/p>\n    <\/div>\n    <div class=\"kpiband four\">\n      <div class=\"kpi\"><div class=\"num sm\">88%<\/div><p>\u00d8 Genauigkeit (cANLS@0.8), h\u00f6chster Wert unter 7 getesteten Modellen<\/p><\/div>\n      <div class=\"kpi\"><div class=\"num sm\">2.0\u00d7<\/div><p>schneller im Schnitt als das genaueste der sechs Referenzmodelle<\/p><\/div>\n      <div class=\"kpi\"><div class=\"num sm\">4<\/div><p>Einsatzoptionen, von On-Prem bis zur API<\/p><\/div>\n      <div class=\"kpi\"><div class=\"num sm\">24 GB<\/div><p>GPU-Speicher gen\u00fcgen f\u00fcr den vollst\u00e4ndigen, air-gapped Betrieb auf einer einzelnen Consumer-Grafikkarte<\/p><\/div>\n    <\/div>\n    <div class=\"note plain\" style=\"margin-top:40px;\">\n      <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"5\" y=\"10.5\" width=\"14\" height=\"10\" rx=\"1.5\"\/><path d=\"M8 10.5V7a4 4 0 1 1 8 0v3.5\"\/><\/svg><\/div>\n      <div><h3 style=\"color:var(--ink)\">Air-gapped-f\u00e4hig<\/h3><p style=\"color:var(--ink-soft)\">Vollst\u00e4ndiger On-Prem-Betrieb auf einer einzelnen Consumer-GPU mit 24 GB VRAM, ohne Rechenzentrum, ohne Cloud-Anbindung.<\/p><\/div>\n    <\/div>\n  <\/div>\n<\/section>\n\n<section class=\"alt\">\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\">Grundlagen<\/span>\n      <h2>Was ist ein Small Special-Purpose Vision Language Model?<\/h2>\n      <p>Ein <b>Vision Language Model (VLM)<\/b> verarbeitet Bild und Text gemeinsam. Es liest eine Dokumentenseite direkt als Bild, inklusive Layout, Tabellen, Stempel und Handschrift, und beantwortet dazu eine textuelle Anfrage. Ein separater OCR-Vorverarbeitungsschritt ist daf\u00fcr nicht erforderlich.<\/p>\n    <\/div>\n    <div class=\"flow\">\n      <div class=\"node\"><div class=\"k\">Eingang<\/div><div class=\"t\">Dokument (Bild)<\/div><span class=\"sub\">Layout, Tabellen, Stempel<\/span><\/div>\n      <div class=\"arrow\">\u2192<\/div>\n      <div class=\"node hl\"><div class=\"k\">Kernmodell<\/div><div class=\"t\">Dokumentli\u00ae VLM<\/div><span class=\"sub\">Bild + Text in einem Schritt<\/span><\/div>\n      <div class=\"arrow\">\u2192<\/div>\n      <div class=\"node\"><div class=\"k\">Ausgabe<\/div><div class=\"t\">Strukturierte Daten<\/div><span class=\"sub\">JSON nach eigenem Schema<\/span><\/div>\n    <\/div>\n    <div class=\"grid3\" style=\"margin-top:44px;\">\n      <div class=\"case\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M7 3h7l4 4v13a1 1 0 0 1-1 1H7a1 1 0 0 1-1-1V4a1 1 0 0 1 1-1Z\"\/><path d=\"M14 3v4h4\"\/><path d=\"M9 12h6\"\/><path d=\"M9 15.5h6\"\/><\/svg><\/div>\n        <h3>Bild + Text in einem Modell<\/h3>\n        <p>Layout, Tabellen und Textinhalt werden gemeinsam interpretiert, ohne separate OCR-Pipeline.<\/p>\n      <\/div>\n      <div class=\"case\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M12 3l9 5-9 5-9-5 9-5Z\"\/><path d=\"M3 13l9 5 9-5\"\/><\/svg><\/div>\n        <h3>Auf Dokumente spezialisiert<\/h3>\n        <p>Trainingsdaten aus echten Gesch\u00e4ftsdokumenten statt allgemeinem Web-Text.<\/p>\n      <\/div>\n      <div class=\"case\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"6\" y=\"6\" width=\"12\" height=\"12\" rx=\"2\"\/><path d=\"M9 3v3M15 3v3M9 18v3M15 18v3M3 9h3M3 15h3M18 9h3M18 15h3\"\/><\/svg><\/div>\n        <h3>Kompakt<\/h3>\n        <p>Weniger Parameter und geringerer Rechenaufwand je Dokument \u2013 l\u00e4uft vollst\u00e4ndig auf einer einzelnen Consumer-GPU mit 24 GB VRAM.<\/p>\n      <\/div>\n    <\/div>\n    <div class=\"kpiband\" style=\"grid-template-columns:1fr 1fr;max-width:680px;margin-top:40px;gap:22px;\">\n      <div class=\"kpi\"><div class=\"num sm\">1 GPU<\/div><p>gen\u00fcgt f\u00fcr den vollst\u00e4ndigen, air-gapped Betrieb \u2013 kein Rechenzentrum, keine Cloud-Anbindung n\u00f6tig.<\/p><\/div>\n      <div class=\"kpi\"><div class=\"num sm\">24 GB<\/div><p>VRAM einer einzelnen Consumer-Grafikkarte reichen f\u00fcr den kompletten Betrieb des Modells.<\/p><\/div>\n    <\/div>\n  <\/div>\n<\/section>\n\n<section class=\"deep\" id=\"benchmarks\">\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\" style=\"background:rgba(73,135,245,.18);color:#7dabf9;border-color:rgba(73,135,245,.3)\">Benchmark-Ergebnisse<\/span>\n      <h2>H\u00f6chste Genauigkeit bei h\u00f6chster Geschwindigkeit<\/h2>\n      <p>In einem unabh\u00e4ngigen Test auf 5 Benchmark-Datens\u00e4tzen mit insgesamt 3'368 realen Gesch\u00e4ftsdokumenten wurde Dokumentli\u00ae gegen sechs aktuelle Referenzmodelle getestet: Claude Opus 4.8, GPT-5.6 Sol, Gemini 3.6 Flash, Gemini 3.7 Flash, Qwen3.6-Plus und das offene Modell Gemma 4 26B-A4B-IT. G\u00fctema\u00df ist cANLS@0.8 (Average Normalized Levenshtein Similarity mit Schwellenwert 0.8), eine g\u00e4ngige Variante des ANLS-Industriestandards f\u00fcr die Bewertung von Dokumenten-Extraktion.<\/p>\n    <\/div>\n\n    <div class=\"barchart-card\">\n      <div class=\"bc-head\">\n        <h3>\u00d8 Genauigkeit (cANLS@0.8) je Modell, gemittelt \u00fcber 5 Benchmarks<\/h3>\n        <span class=\"bc-flag\">Mehr Genauigkeit \u2191<\/span>\n      <\/div>\n      <div class=\"barchart\">\n        <div class=\"bar best\"><span class=\"crown\">Bestwert<\/span><span class=\"val\">88.4%<\/span><div class=\"fill\" style=\"height:135px\"><\/div><span class=\"lbl\">Dokumentli\u00ae<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">83.5%<\/span><div class=\"fill\" style=\"height:112px\"><\/div><span class=\"lbl\">Gemini 3.6<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">82.9%<\/span><div class=\"fill\" style=\"height:109px\"><\/div><span class=\"lbl\">Gemini 3.7<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">81.8%<\/span><div class=\"fill\" style=\"height:104px\"><\/div><span class=\"lbl\">GPT-5.6 Sol<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">82.1%<\/span><div class=\"fill\" style=\"height:105px\"><\/div><span class=\"lbl\">Qwen3.6<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">81.0%<\/span><div class=\"fill\" style=\"height:100px\"><\/div><span class=\"lbl\">Claude 4.8<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">69.0%<\/span><div class=\"fill\" style=\"height:43px\"><\/div><span class=\"lbl\">Gemma 4 26B<\/span><\/div>\n      <\/div>\n      <p class=\"bc-axisnote\">Skala beginnt bei 60% cANLS@0.8 zur besseren Lesbarkeit der Unterschiede, exakte Werte siehe Balkenbeschriftung.<\/p>\n    <\/div>\n\n    <div class=\"barchart-card\">\n      <div class=\"bc-head\">\n        <h3>\u00d8 Verarbeitungsgeschwindigkeit je Modell, Dokumente pro Sekunde<\/h3>\n        <span class=\"bc-flag\">Mehr Tempo \u2191<\/span>\n      <\/div>\n      <div class=\"barchart\">\n        <div class=\"bar best\"><span class=\"crown\">Bestwert<\/span><span class=\"val\">0.86<\/span><div class=\"fill\" style=\"height:130px\"><\/div><span class=\"lbl\">Dokumentli\u00ae<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">0.82<\/span><div class=\"fill\" style=\"height:124px\"><\/div><span class=\"lbl\">Gemini 3.7<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">0.80<\/span><div class=\"fill\" style=\"height:121px\"><\/div><span class=\"lbl\">Claude 4.8<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">0.43<\/span><div class=\"fill\" style=\"height:65px\"><\/div><span class=\"lbl\">Gemini 3.6<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">0.34<\/span><div class=\"fill\" style=\"height:51px\"><\/div><span class=\"lbl\">Gemma 4 26B<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">0.32<\/span><div class=\"fill\" style=\"height:48px\"><\/div><span class=\"lbl\">GPT-5.6 Sol<\/span><\/div>\n        <div class=\"bar\"><span class=\"val\">0.04<\/span><div class=\"fill\" style=\"height:6px\"><\/div><span class=\"lbl\">Qwen3.6<\/span><\/div>\n      <\/div>\n    <\/div>\n\n    <div class=\"grid3 compliance\" style=\"grid-template-columns:1fr 1fr;margin-top:22px;\">\n      <div class=\"case\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M9 12.5l2 2 4-4.5\"\/><circle cx=\"12\" cy=\"12\" r=\"9\"\/><\/svg><\/div>\n        <h3>5 \/ 5<\/h3>\n        <p>Benchmarks, in denen Dokumentli\u00ae vor Gemini 3.6 Flash liegt, dem st\u00e4rksten der sechs Referenzmodelle.<\/p>\n      <\/div>\n      <div class=\"case\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"6\" cy=\"6\" r=\"2.2\"\/><circle cx=\"18\" cy=\"6\" r=\"2.2\"\/><circle cx=\"12\" cy=\"18\" r=\"2.2\"\/><path d=\"M7.8 7.2L11 16M16.2 7.2L13 16\"\/><\/svg><\/div>\n        <h3>7<\/h3>\n        <p>getestete Modelle insgesamt, Dokumentli\u00ae plus sechs general-purpose Frontier-Modelle als Referenz.<\/p>\n      <\/div>\n    <\/div>\n    <div class=\"note deepbox\" style=\"margin-top:22px;\">\n      <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M7 3h7l4 4v13a1 1 0 0 1-1 1H7a1 1 0 0 1-1-1V4a1 1 0 0 1 1-1Z\"\/><path d=\"M14 3v4h4\"\/><path d=\"M9 12h6\"\/><path d=\"M9 15.5h6\"\/><\/svg><\/div>\n      <div><h3>F\u00fcr Details siehe den Benchmark Report<\/h3><p>Ergebnisse je Branche und Modell, Konfidenzintervalle und Positionierungs-Matrizen sowie die vollst\u00e4ndige Methodik sind im separaten <b style=\"color:var(--invert)\">Parashift Dokumentli\u00ae Benchmark Report<\/b> dokumentiert.<\/p><\/div>\n    <\/div>\n  <\/div>\n<\/section>\n\n<section>\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\">Spezialisierung<\/span>\n      <h2>Warum ein spezialisiertes Modell general-purpose LLMs schl\u00e4gt<\/h2>\n      <p>General-purpose Frontier-Modelle sind darauf trainiert, ein sehr breites Aufgabenspektrum abzudecken. F\u00fcr die eng definierte Aufgabe \u201elies dieses Gesch\u00e4ftsdokument und extrahiere diese Felder\" ist das ein Trainingsziel, das an der eigentlichen Aufgabe vorbeizielt. Vier Faktoren erkl\u00e4ren, warum ein kleineres, spezialisiertes Modell hier in den Benchmarks vorne liegt:<\/p>\n    <\/div>\n    <div class=\"usecase-feature\">\n      <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M12 3l9 5-9 5-9-5 9-5Z\"\/><path d=\"M3 13l9 5 9-5\"\/><\/svg><\/div>\n      <div>\n        <span class=\"kicker\">Trainingsdaten-Fokus<\/span>\n        <h3>Millionen reale Gesch\u00e4ftsdokumente statt allgemeinem Web-Text<\/h3>\n        <p>Inklusive der Eigenheiten von Rechnungen, Frachtpapieren und Vertr\u00e4gen aus regulierten Branchen.<\/p>\n      <\/div>\n    <\/div>\n    <div class=\"usecase-feature\">\n      <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M7 3h7l4 4v13a1 1 0 0 1-1 1H7a1 1 0 0 1-1-1V4a1 1 0 0 1 1-1Z\"\/><path d=\"M14 3v4h4\"\/><path d=\"M12 11v5\"\/><path d=\"M9.5 13.5h5\"\/><\/svg><\/div>\n      <div>\n        <span class=\"kicker\">Architektur f\u00fcr Layout-Verst\u00e4ndnis<\/span>\n        <h3>Bild und Text werden in einem Schritt verarbeitet<\/h3>\n        <p>Ohne verlustbehaftete OCR-Vorstufe, relevant bei Tabellen, mehrspaltigem Layout, Stempeln und handschriftlichen Vermerken.<\/p>\n      <\/div>\n    <\/div>\n    <div class=\"usecase-feature\">\n      <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"12\" cy=\"12\" r=\"8.5\"\/><circle cx=\"12\" cy=\"12\" r=\"5\"\/><circle cx=\"12\" cy=\"12\" r=\"1.4\" fill=\"currentColor\" stroke=\"none\"\/><\/svg><\/div>\n      <div>\n        <span class=\"kicker\">Geringerer Kontextbedarf<\/span>\n        <h3>Ein kleineres Kontextbudget je Anfrage<\/h3>\n        <p>Das senkt Latenz und Rechenaufwand pro Dokument sp\u00fcrbar \u2013 und macht den Betrieb auf einer einzelnen Consumer-GPU mit 24 GB VRAM erst m\u00f6glich.<\/p>\n      <\/div>\n    <\/div>\n    <div class=\"usecase-feature\" style=\"margin-bottom:0;\">\n      <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M4 6h9M17 6h3M4 12h3M11 12h9M4 18h13\"\/><circle cx=\"15\" cy=\"6\" r=\"2\"\/><circle cx=\"7\" cy=\"12\" r=\"2\"\/><circle cx=\"16\" cy=\"18\" r=\"2\"\/><\/svg><\/div>\n      <div>\n        <span class=\"kicker\">Nachtrainierbar auf eigene Dokumenttypen <span class=\"tag-soon\">Demn\u00e4chst verf\u00fcgbar<\/span><\/span>\n        <h3>Kontinuierliches Feintuning statt nur Prompt-Anpassung<\/h3>\n        <p>Feintuning auf unternehmensspezifische Vorlagen und Sonderf\u00e4lle statt nur generischer Prompt-Anpassung.<\/p>\n      <\/div>\n    <\/div>\n    <div class=\"genspec\">\n      <div>\n        <h4>Generalist (general-purpose LLM)<\/h4>\n        <p>Ein Modell, viele m\u00f6gliche Aufgaben: Konversation, Code, Weltwissen, Dokumente. Dokumentenverst\u00e4ndnis ist eine von vielen F\u00e4higkeiten, nicht der Trainingsschwerpunkt.<\/p>\n      <\/div>\n      <div class=\"spec\">\n        <h4>Spezialist (Dokumentli\u00ae)<\/h4>\n        <p>Ein Modell, eine Aufgabe: Gesch\u00e4ftsdokumente lesen, klassifizieren, strukturieren. Trainingsdaten, Architektur und Kontextbudget sind vollst\u00e4ndig darauf ausgerichtet.<\/p>\n      <\/div>\n    <\/div>\n  <\/div>\n<\/section>\n\n<section class=\"alt\">\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\">Abgrenzung<\/span>\n      <h2>Was Dokumentli\u00ae nicht kann<\/h2>\n      <p>Die Spezialisierung, die Dokumentli\u00ae bei Dokumenten-Intelligence-Aufgaben stark macht, ist zugleich eine bewusste Grenze: Dokumentli\u00ae ist kein general-purpose Assistent. Ausserhalb der Dokumentenverarbeitung ist ein general-purpose Modell oder, noch besser, ein auf die jeweilige Automatisierungsaufgabe spezialisiertes Modell die richtige Wahl.<\/p>\n    <\/div>\n    <div class=\"tablewrap\">\n      <table class=\"cmp\">\n        <thead>\n          <tr>\n            <th style=\"min-width:260px\">Aufgabe<\/th>\n            <th class=\"hl\">Dokumentli\u00ae<\/th>\n            <th>General-purpose Frontier-Modell<\/th>\n          <\/tr>\n        <\/thead>\n        <tbody>\n          <tr>\n            <td class=\"feat\"><b>Dokumente klassifizieren &amp; Felder extrahieren<\/b><\/td>\n            <td><span class=\"mark yes\">\u2713 Kernaufgabe, daf\u00fcr trainiert<\/span><\/td>\n            <td><span class=\"mark no\">\u2715 M\u00f6glich<\/span><span class=\"riskchip-note\">mit geringerer Genauigkeit und Geschwindigkeit, h\u00f6heren Kosten und h\u00f6herem Sicherheits-\/Compliance-Aufwand<\/span><\/td>\n          <\/tr>\n          <tr>\n            <td class=\"feat\"><b>Freies Reasoning &amp; offene Konversation<\/b><\/td>\n            <td><span class=\"mark no\">\u2715 Nicht vorgesehen<\/span><\/td>\n            <td><span class=\"mark plain\">\u2713 Kernaufgabe<\/span><\/td>\n          <\/tr>\n          <tr>\n            <td class=\"feat\"><b>Programmierung \/ Code-Generierung<\/b><\/td>\n            <td><span class=\"mark no\">\u2715 Nicht vorgesehen<\/span><\/td>\n            <td><span class=\"mark plain\">\u2713 Kernaufgabe<\/span><\/td>\n          <\/tr>\n          <tr>\n            <td class=\"feat\"><b>Allgemeine Wissensfragen ausserhalb des Dokumentinhalts<\/b><\/td>\n            <td><span class=\"mark no\">\u2715 Nicht vorgesehen<\/span><\/td>\n            <td><span class=\"mark plain\">\u2713 Kernaufgabe<\/span><\/td>\n          <\/tr>\n          <tr>\n            <td class=\"feat\"><b>Freies kreatives Schreiben<\/b><\/td>\n            <td><span class=\"mark no\">\u2715 Nicht vorgesehen<\/span><\/td>\n            <td><span class=\"mark plain\">\u2713 Kernaufgabe<\/span><\/td>\n          <\/tr>\n          <tr>\n            <td class=\"feat\"><b>Mehrstufige, dokumentfremde Agenten-Aufgaben<\/b><\/td>\n            <td><span class=\"mark no\">\u2715 Nicht vorgesehen<\/span><\/td>\n            <td><span class=\"mark plain\">\u2713 M\u00f6glich<\/span><\/td>\n          <\/tr>\n        <\/tbody>\n      <\/table>\n    <\/div>\n    <div class=\"decision-grid\" style=\"margin-top:28px;\">\n      <div class=\"decision yes\">\n        <div class=\"dh\">\n          <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"12\" cy=\"12\" r=\"9\"\/><path d=\"M8 12.5l2.5 2.5L16 9.5\"\/><\/svg><\/div>\n          <h3>Dokumentli\u00ae einsetzen, wenn\u2026<\/h3>\n        <\/div>\n        <ul>\n          <li>hohe Volumen an Gesch\u00e4ftsdokumenten klassifiziert oder Felder daraus extrahiert werden m\u00fcssen<\/li>\n          <li>Datenhoheit, Air-Gapped-Betrieb oder planbare Kosten wichtig sind<\/li>\n          <li>strukturierte, schemakonforme Ausgaben statt freier Text ben\u00f6tigt werden<\/li>\n        <\/ul>\n      <\/div>\n      <div class=\"decision no\">\n        <div class=\"dh\">\n          <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"6\" cy=\"6\" r=\"2.2\"\/><circle cx=\"18\" cy=\"6\" r=\"2.2\"\/><circle cx=\"12\" cy=\"18\" r=\"2.2\"\/><path d=\"M7.8 7.2L11 16M16.2 7.2L13 16\"\/><\/svg><\/div>\n          <h3>General-purpose LLM einsetzen, wenn\u2026<\/h3>\n        <\/div>\n        <ul>\n          <li>die Aufgabe offenes Reasoning, Konversation oder Code betrifft<\/li>\n          <li>allgemeines Weltwissen ausserhalb des Dokumentinhalts gefragt ist<\/li>\n          <li>es um freies kreatives Schreiben oder dokumentfremde Agenten-Aufgaben geht<\/li>\n        <\/ul>\n      <\/div>\n    <\/div>\n    <div class=\"note plain\" style=\"margin-top:22px;\">\n      <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"6\" cy=\"6\" r=\"2.2\"\/><circle cx=\"18\" cy=\"6\" r=\"2.2\"\/><circle cx=\"12\" cy=\"18\" r=\"2.2\"\/><path d=\"M7.8 7.2L11 16M16.2 7.2L13 16\"\/><\/svg><\/div>\n      <div><h3 style=\"color:var(--ink)\">Kombinierbar statt exklusiv<\/h3><p style=\"color:var(--ink-soft)\">Unternehmen setzen Dokumentli\u00ae gezielt f\u00fcr die Dokumenteneingangsverarbeitung ein und kombinieren es bei Bedarf mit general-purpose Modellen f\u00fcr nachgelagerte, dokumentfremde Schritte. Beides l\u00e4sst sich \u00fcber die eigene Workflow-L\u00f6sung orchestrieren.<\/p><\/div>\n    <\/div>\n  <\/div>\n<\/section>\n\n<section class=\"deep\">\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\" style=\"background:rgba(73,135,245,.18);color:#7dabf9;border-color:rgba(73,135,245,.3)\">Einsatzoptionen<\/span>\n      <h2>Vier Wege, Dokumentli\u00ae zu betreiben<\/h2>\n      <p>Dokumentli\u00ae l\u00e4sst sich je nach Datenhoheits-, Compliance- und Integrationsanforderung auf vier Arten betreiben, vom vollst\u00e4ndig isolierten Betrieb im eigenen Rechenzentrum bis zur direkten API-Einbindung.<\/p>\n    <\/div>\n    <div class=\"note deepbox\" style=\"margin-bottom:28px;\">\n      <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M12 3l9 5-9 5-9-5 9-5Z\"\/><path d=\"M3 13l9 5 9-5\"\/><\/svg><\/div>\n      <div><h3>Das Modell und die Node-Komponente<\/h3><p>Dokumentli\u00ae ist das KI-Modell selbst. F\u00fcr den einfachen Betrieb auf eigener Hardware liefert Parashift zus\u00e4tzlich den <b style=\"color:var(--invert)\">Dokumentli\u00ae Node<\/b>, eine leichtgewichtige Komponente, die das Modell verpackt und die Einrichtung on-prem oder air-gapped in kurzer Zeit erm\u00f6glicht, ohne dass ein eigenes Deployment-Setup gebaut werden muss.<\/p><\/div>\n    <\/div>\n    <div class=\"grid4\">\n      <div class=\"deploy-card hl\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"4\" y=\"4\" width=\"16\" height=\"6\" rx=\"1.2\"\/><rect x=\"4\" y=\"14\" width=\"16\" height=\"6\" rx=\"1.2\"\/><circle cx=\"7.5\" cy=\"7\" r=\".9\" fill=\"currentColor\" stroke=\"none\"\/><circle cx=\"7.5\" cy=\"17\" r=\".9\" fill=\"currentColor\" stroke=\"none\"\/><\/svg><\/div>\n        <h3>On-Prem \/ eigene Hardware<\/h3>\n        <span class=\"dtag\">Maximale Datenhoheit<\/span>\n        <ul>\n          <li><b>Betriebsort:<\/b> eigenes Rechenzentrum, via Dokumentli\u00ae Node<\/li>\n          <li><b>Air-Gapped-f\u00e4hig:<\/b> keine Daten verlassen das eigene Netzwerk<\/li>\n          <li><b>Hardware:<\/b> eine konsumenten-taugliche GPU mit 24 GB VRAM reicht<\/li>\n          <li><b>Typisch f\u00fcr:<\/b> h\u00f6chste Compliance-Anforderungen, keine Internetanbindung n\u00f6tig<\/li>\n        <\/ul>\n      <\/div>\n      <div class=\"deploy-card\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M7 18a4.5 4.5 0 0 1-.5-8.97A5.5 5.5 0 0 1 17.2 8.2 4 4 0 0 1 16.5 18H7Z\"\/><\/svg><\/div>\n        <h3>Private-Cloud-Instanz<\/h3>\n        <span class=\"dtag\">Dedizierter Tenant<\/span>\n        <ul>\n          <li><b>Betriebsort:<\/b> isolierte Cloud-Instanz in Region\/Zone der Wahl<\/li>\n          <li><b>Datenhoheit:<\/b> sehr hoch, dedizierter Tenant<\/li>\n          <li><b>Hardware:<\/b> vom Cloud-Anbieter bereitgestellt, nicht von Parashift<\/li>\n          <li><b>Typisch f\u00fcr:<\/b> regulierte Unternehmen mit eigener Cloud-Strategie<\/li>\n        <\/ul>\n      <\/div>\n      <div class=\"deploy-card\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M12 3l9 5-9 5-9-5 9-5Z\"\/><path d=\"M3 13l9 5 9-5\"\/><\/svg><\/div>\n        <h3>Parashift Plattform<\/h3>\n        <span class=\"dtag\">Demn\u00e4chst verf\u00fcgbar<\/span>\n        <ul>\n          <li>Document Intelligence AI Guardrail Platform<\/li>\n          <li><b>Betriebsort:<\/b> Parashifts europ\u00e4ische Sovereign-Cloud-Plattform<\/li>\n          <li><b>Datenhoheit:<\/b> hoch, Compliance-Zonen EU, Schweiz, Deutschland<\/li>\n          <li><b>Zusatzfunktionen:<\/b> Ingestion, Separation, Validierung, Automatisierung, 30+ Konnektoren<\/li>\n          <li><b>Typisch f\u00fcr:<\/b> Teams, die die gesamte Dokumenten-Pipeline nutzen m\u00f6chten<\/li>\n        <\/ul>\n      <\/div>\n      <div class=\"deploy-card\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M8 4L4 12l4 8\"\/><path d=\"M16 4l4 8-4 8\"\/><\/svg><\/div>\n        <h3>Dokumentli\u00ae API<\/h3>\n        <span class=\"dtag\">Demn\u00e4chst verf\u00fcgbar<\/span>\n        <ul>\n          <li>Direkter Modell-Cloud-API-Zugriff<\/li>\n          <li><b>Betriebsort:<\/b> programmatischer Endpunkt, ohne Plattform-Oberfl\u00e4che<\/li>\n          <li><b>Hardware:<\/b> keine, vollst\u00e4ndig verwaltet<\/li>\n          <li><b>Flexibel:<\/b> gegen eigenen Dokumentli-Node- oder Parashift-Endpunkt<\/li>\n          <li><b>Typisch f\u00fcr:<\/b> Teams, die Dokumentli\u00ae in eine eigene Applikation einbetten<\/li>\n        <\/ul>\n      <\/div>\n    <\/div>\n    <div class=\"note brand\" style=\"margin-top:28px;\">\n      <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M14.7 6.3a1 1 0 0 0 0 1.4l1.6 1.6a1 1 0 0 0 1.4 0l3.4-3.4a6 6 0 0 1-7.9 7.9L4.9 22 2 19.1l8.3-8.3a6 6 0 0 1 7.9-7.9l-3.4 3.4Z\"\/><\/svg><\/div>\n      <div><span class=\"kicker\">Anbieteroffen<\/span><h3>Offen f\u00fcr jede Agentic-Automatisierungs- und Workflow-L\u00f6sung<\/h3><p>On-Prem, Private-Cloud-Instanz und die Dokumentli\u00ae API lassen sich mit praktisch jeder Dokumenten- oder Workflow-Automatisierungsl\u00f6sung am Markt verbinden, ob RPA, iPaaS, ERP-, DMS- oder eine selbst entwickelte Applikation. Voraussetzung ist lediglich, dass die jeweilige L\u00f6sung einen API-Aufruf ausf\u00fchren kann, eine Bindung an eine bestimmte Plattform ist nicht erforderlich.<\/p><\/div>\n    <\/div>\n  <\/div>\n<\/section>\n\n<section>\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\">Nutzung &amp; Integration<\/span>\n      <h2>Passt sich in Ihren bestehenden Stack ein oder in die Parashift Plattform<\/h2>\n      <p style=\"margin-top:20px;\">Dokumentli\u00ae ist bewusst so gebaut, dass es sich unabh\u00e4ngig von der gew\u00e4hlten Einsatzoption in praktisch jede bestehende Dokumenten- oder Workflow-Automatisierungslandschaft einf\u00fcgt. Zus\u00e4tzlich l\u00e4sst sich Dokumentli\u00ae auf zwei grunds\u00e4tzliche Arten einbinden.<\/p>\n    <\/div>\n    <div class=\"decision-grid\">\n      <div class=\"decision plain\">\n        <div class=\"miniflow\">\n          <div class=\"mf-node\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"5\" y=\"4\" width=\"14\" height=\"17\" rx=\"1.5\"\/><\/svg><\/div>\n          <span class=\"mf-arrow\">\u2192<\/span>\n          <div class=\"mf-node\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M8 4L4 12l4 8\"\/><path d=\"M16 4l4 8-4 8\"\/><\/svg><\/div>\n          <span class=\"mf-arrow\">\u2192<\/span>\n          <div class=\"mf-node\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"6\" y=\"6\" width=\"12\" height=\"12\" rx=\"2\"\/><\/svg><\/div>\n          <span class=\"mf-arrow\">\u2192<\/span>\n          <div class=\"mf-node\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"3.5\" y=\"5.5\" width=\"17\" height=\"13\" rx=\"1.5\"\/><\/svg><\/div>\n        <\/div>\n        <span class=\"kicker\">Direkte Einbindung<\/span>\n        <h3>Programmatisch, in jede bestehende L\u00f6sung<\/h3>\n        <ul class=\"plain\">\n          <li><b>Volle Kontrolle:<\/b> Dokumentli\u00ae als Baustein in Ihrem ERP, DMS, Ihrer RPA-\/iPaaS-L\u00f6sung oder eigenen Applikation.<\/li>\n          <li><b>Eigenes Schema:<\/b> die Anfrage gibt das gew\u00fcnschte Datenformat vor.<\/li>\n          <li><b>Anbieteroffen:<\/b> keine Bindung an eine bestimmte Plattform, nur ein API-Aufruf n\u00f6tig.<\/li>\n        <\/ul>\n      <\/div>\n      <div class=\"decision plain\">\n        <div class=\"miniflow\">\n          <div class=\"mf-node\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M7 3h7l4 4v13a1 1 0 0 1-1 1H7a1 1 0 0 1-1-1V4a1 1 0 0 1 1-1Z\"\/><\/svg><\/div>\n          <span class=\"mf-arrow\">\u2192<\/span>\n          <div class=\"mf-node\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M12 3l9 5-9 5-9-5 9-5Z\"\/><path d=\"M3 13l9 5 9-5\"\/><\/svg><\/div>\n          <span class=\"mf-arrow\">\u2192<\/span>\n          <div class=\"mf-node\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"6\" y=\"6\" width=\"12\" height=\"12\" rx=\"2\"\/><\/svg><\/div>\n          <span class=\"mf-arrow\">\u2192<\/span>\n          <div class=\"mf-node\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"12\" cy=\"12\" r=\"9\"\/><path d=\"M8 12.5l2.5 2.5L16 9.5\"\/><\/svg><\/div>\n          <span class=\"mf-arrow\">\u2192<\/span>\n          <div class=\"mf-node\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M4 12h4l2 3h4l2-3h4\"\/><path d=\"M4 12V6a1 1 0 0 1 1-1h14a1 1 0 0 1 1 1v6\"\/><path d=\"M4 12v6a1 1 0 0 0 1 1h14a1 1 0 0 0 1-1v-6\"\/><\/svg><\/div>\n        <\/div>\n        <span class=\"kicker\">Optional: Workflow-L\u00f6sung<\/span>\n        <h3>Zus\u00e4tzlich orchestriert \u00fcber die Parashift Plattform<\/h3>\n        <ul class=\"plain\">\n          <li><b>F\u00fcr alle, die keine eigene Orchestrierung betreiben m\u00f6chten:<\/b> eingebettete Validierungs-Oberfl\u00e4che, Human-in-the-Loop ohne eigene Frontend-Entwicklung.<\/li>\n          <li><b>30+ Konnektoren:<\/b> ERP-, DMS-, CRM-Systeme und E-Mail, ohne eigene Middleware.<\/li>\n          <li><b>Governance inklusive:<\/b> Konfidenzwerte, Routing-Schwellen, Audit-Trail, 2-\/3-Way-Match.<\/li>\n        <\/ul>\n      <\/div>\n    <\/div>\n    <p style=\"margin-top:32px;color:var(--ink-soft);font-size:15.5px;max-width:76ch;\">In der Praxis ist die Wahl keine Entweder-Oder-Entscheidung: bestehende Dokumenten- und Workflow-Automatisierungsl\u00f6sungen binden Dokumentli\u00ae direkt \u00fcber die API ein, w\u00e4hrend Teams ohne eigene Orchestrierung zus\u00e4tzlich die vorgefertigte Parashift Plattform nutzen k\u00f6nnen. Beide Muster lassen sich auf derselben Dokumentli\u00ae-Instanz kombinieren.<\/p>\n    <p class=\"intchain\">RPA \u00b7 iPaaS \u00b7 ERP \u00b7 DMS \u00b7 CRM \u00b7 E-Mail-Postf\u00e4cher \u00b7 Eigene Applikationen<\/p>\n  <\/div>\n<\/section>\n\n<section class=\"alt\">\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\">Kostenlogik<\/span>\n      <h2>Warum Dokumentli\u00ae g\u00fcnstiger ist als general-purpose LLM-APIs<\/h2>\n      <p>General-purpose Frontier-Modelle werden bei Cloud-Anbietern in aller Regel pro API-Aufruf und verarbeitetem Token abgerechnet. F\u00fcr Unternehmen mit hohem und schwankendem Dokumentenvolumen ergeben sich daraus zwei strukturelle Nachteile: Die Kosten skalieren direkt mit dem Volumen, ohne planbare Obergrenze, und zus\u00e4tzlich mit dem Kontextbudget, das ein general-purpose Modell je Anfrage ben\u00f6tigt.<\/p>\n    <\/div>\n    <div class=\"costcard\">\n      <div class=\"cc-head\">\n        <h3>Kostenverlauf \u00fcber die Zeit, schematische Darstellung<\/h3>\n        <p>Illustrativ, ohne konkrete Betr\u00e4ge: Pay-per-Call-Kosten schwanken mit Volumen und Nachfrage-Spitzen; ein Seitenkontingent-Abonnement bleibt konstant und planbar.<\/p>\n      <\/div>\n      <div class=\"costlegend\">\n        <span><i class=\"ppc\"><\/i>Pay-per-Call (General-purpose LLM-API)<\/span>\n        <span><i class=\"sub\"><\/i>Dokumentli\u00ae (Abonnement)<\/span>\n        <span class=\"cost-flag\">Kosten, Achse illustrativ \u2193<\/span>\n      <\/div>\n      <div class=\"costchart\">\n        <div class=\"cgroup\"><div class=\"cbar ppc\" style=\"height:70px\"><\/div><div class=\"cbar sub\" style=\"height:88px\"><\/div><\/div>\n        <div class=\"cgroup\"><div class=\"cbar ppc\" style=\"height:115px\"><\/div><div class=\"cbar sub\" style=\"height:88px\"><\/div><\/div>\n        <div class=\"cgroup\"><div class=\"cbar ppc\" style=\"height:45px\"><\/div><div class=\"cbar sub\" style=\"height:88px\"><\/div><\/div>\n        <div class=\"cgroup\"><div class=\"cbar ppc\" style=\"height:138px\"><\/div><div class=\"cbar sub\" style=\"height:88px\"><\/div><\/div>\n        <div class=\"cgroup\"><div class=\"cbar ppc\" style=\"height:80px\"><\/div><div class=\"cbar sub\" style=\"height:88px\"><\/div><\/div>\n        <div class=\"cgroup\"><div class=\"cbar ppc\" style=\"height:122px\"><\/div><div class=\"cbar sub\" style=\"height:88px\"><\/div><\/div>\n      <\/div>\n      <div class=\"costaxis\">\n        <span>Monat 1<\/span><span>Monat 2<\/span><span>Monat 3<\/span><span>Monat 4<\/span><span>Monat 5<\/span><span>Monat 6<\/span>\n      <\/div>\n    <\/div>\n    <div class=\"grid3\" style=\"grid-template-columns:1fr 1fr;margin-top:22px;\">\n      <div class=\"case\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"9\" cy=\"9\" r=\"5.5\"\/><circle cx=\"15\" cy=\"15\" r=\"5.5\"\/><\/svg><\/div>\n        <h3>Pay-per-Call<\/h3>\n        <p>General-purpose LLM-APIs: Kosten steigen direkt und ohne Obergrenze mit Volumen und Tokenverbrauch.<\/p>\n      <\/div>\n      <div class=\"case\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M12 3l7 3.5v5c0 4.8-3 8.6-7 9.5-4-.9-7-4.7-7-9.5v-5L12 3Z\"\/><path d=\"M9 12.2l2.1 2.1L15.5 10\"\/><\/svg><\/div>\n        <h3>Kapazit\u00e4tsbasiert<\/h3>\n        <p>Dokumentli\u00ae: Abonnement mit Seitenkontingent, im Voraus planbar, unabh\u00e4ngig von Nachfrage-Spitzen.<\/p>\n      <\/div>\n    <\/div>\n    <p style=\"margin-top:28px;color:var(--ink-soft);font-size:15.5px;max-width:76ch;\">Objektivierbar wird das \u00fcber den geringeren Rechenaufwand pro Dokument: Dokumentli\u00ae verarbeitet ein Dokument mit sp\u00fcrbar weniger Rechenleistung als general-purpose Referenzmodelle, bei gleichzeitig h\u00f6herer Genauigkeit \u2013 und l\u00e4uft wahlweise vollst\u00e4ndig auf einer einzelnen Consumer-GPU mit <b class=\"mono\" style=\"color:var(--ink)\">24 GB<\/b> VRAM statt auf Cloud-Rechenzentren. Das macht die Kosten unabh\u00e4ngig vom Preismodell einzelner Cloud-Anbieter von Grund auf planbar.<\/p>\n    <div class=\"note plain\" style=\"margin-top:22px;\">\n      <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"9\" cy=\"9\" r=\"5.5\"\/><circle cx=\"15\" cy=\"15\" r=\"5.5\"\/><\/svg><\/div>\n      <div><h3 style=\"color:var(--ink)\">Subscriptions<\/h3><p style=\"color:var(--ink-soft)\">Dokumentli\u00ae wird in mehreren Subscriptions angeboten, gestaffelt nach monatlichem Dokumentenvolumen, von Pilotprojekten bis zum Hochvolumen-Einsatz im Unternehmen. Jede Stufe umfasst <b style=\"color:var(--ink)\">4 Basismodell-Updates pro Jahr<\/b>, sodass Genauigkeits- und Funktionsverbesserungen automatisch einfliessen, ohne neuen Vertrag.<\/p><\/div>\n    <\/div>\n  <\/div>\n<\/section>\n\n<section class=\"deep\">\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\" style=\"background:rgba(73,135,245,.18);color:#7dabf9;border-color:rgba(73,135,245,.3)\">Trust &amp; Enterprise Readiness<\/span>\n      <h2>Vertrauen, das \u00fcber Compliance hinausgeht<\/h2>\n    <\/div>\n    <div class=\"grid3 compliance\">\n      <div class=\"case\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"5\" y=\"10.5\" width=\"14\" height=\"10\" rx=\"1.5\"\/><path d=\"M8 10.5V7a4 4 0 1 1 8 0v3.5\"\/><\/svg><\/div>\n        <h3>Data Sovereignty<\/h3>\n        <p>Betrieb On-Prem, air-gapped oder in dedizierten Compliance-Zonen f\u00fcr EU, Schweiz und Deutschland, Ihre Dokumentdaten verlassen die gew\u00e4hlte Zone nie.<\/p>\n      <\/div>\n      <div class=\"case\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M3 3l18 18\"\/><path d=\"M10.6 5.2A10.8 10.8 0 0 1 12 5c5 0 9 4 10.5 7-.6 1.2-1.5 2.5-2.7 3.7M6.1 6.9C4 8.3 2.4 10.2 1.5 12c1.5 3 5.5 7 10.5 7 1.4 0 2.7-.3 3.9-.8\"\/><path d=\"M9.9 10a3 3 0 0 0 4.2 4.2\"\/><\/svg><\/div>\n        <h3>No Shadow AI<\/h3>\n        <p>Dedizierte Tenants statt geteilter Infrastruktur: Ihre vertraulichen Dokumente fliessen nie in ein \u00f6ffentlich zug\u00e4ngliches Modelltraining ein.<\/p>\n      <\/div>\n      <div class=\"case\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M9 2v5\"\/><path d=\"M15 2v5\"\/><path d=\"M6.5 7h11a1.5 1.5 0 0 1 1.5 1.5V11a5 5 0 0 1-5 5h-4a5 5 0 0 1-5-5V8.5A1.5 1.5 0 0 1 6.5 7Z\"\/><path d=\"M12 16v3a3 3 0 0 1-3 3\"\/><\/svg><\/div>\n        <h3>Governance &amp; Audit-Trail<\/h3>\n        <p>Konfidenzwerte, Routing-Schwellen und l\u00fcckenlose Audit-Trails auf der Parashift Plattform, f\u00fcr nachvollziehbare Entscheidungen in jedem Prozessschritt.<\/p>\n      <\/div>\n    <\/div>\n    <div class=\"compliance-badges\">\n      <span class=\"badges-label\">Zertifizierungen &amp; Audits<\/span>\n      <div class=\"badges-row\">\n        <span class=\"badge-logo\"><img decoding=\"async\" src=\"https:\/\/parashift.ai\/wp-content\/uploads\/2025\/11\/ISO27001-certified-500x500-1.png\" alt=\"ISO 27001\"><\/span>\n        <span class=\"badge-logo\"><img decoding=\"async\" src=\"https:\/\/parashift.ai\/wp-content\/uploads\/2025\/11\/AICPA-SOC-2-500x500-1.png\" alt=\"SOC 2\"><\/span>\n        <span class=\"badge-logo\"><img decoding=\"async\" src=\"https:\/\/parashift.ai\/wp-content\/uploads\/2025\/11\/Cloud-Computing-C5-500x500-1.png\" alt=\"BSI C5\"><\/span>\n        <span class=\"badge-logo\"><img decoding=\"async\" src=\"https:\/\/parashift.ai\/wp-content\/uploads\/2025\/11\/PCI-DSS-Compliant-500x500-1.avif\" alt=\"PCI DSS\"><\/span>\n      <\/div>\n    <\/div>\n  <\/div>\n<\/section>\n\n<section>\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\">Fazit &amp; Quellen<\/span>\n      <h2>Ein spezialisiertes Modell f\u00fcr eine klar definierte Aufgabe<\/h2>\n    <\/div>\n    <div class=\"usecase-feature\" style=\"margin-bottom:0;background:var(--bg-deep);border:none;\">\n      <div style=\"width:100%;\">\n        <p style=\"color:var(--invert-soft);font-size:16px;max-width:76ch;margin-bottom:24px;\">Dokumentli\u00ae erreicht die h\u00f6chste durchschnittliche Extraktionsgenauigkeit unter sieben getesteten Modellen und ist zugleich im Durchschnitt das schnellste Modell im Feld \u2013 bei deutlich geringerem Rechenaufwand und wahlweise vollst\u00e4ndig air-gapped auf einer einzelnen Consumer-GPU mit 24 GB VRAM.<\/p>\n        <div class=\"grid4\">\n          <div class=\"deploy-card\" style=\"padding:20px;\">\n            <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"12\" cy=\"12\" r=\"9\"\/><path d=\"M8 12.5l2.5 2.5L16 9.5\"\/><\/svg><\/div>\n            <h3 style=\"font-size:15px;margin-bottom:4px;\">Genauer<\/h3>\n            <p style=\"color:var(--invert-soft);font-size:13.5px;\">als general-purpose LLMs<\/p>\n          <\/div>\n          <div class=\"deploy-card\" style=\"padding:20px;\">\n            <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"6\" y=\"6\" width=\"12\" height=\"12\" rx=\"2\"\/><path d=\"M9 3v3M15 3v3M9 18v3M15 18v3M3 9h3M3 15h3M18 9h3M18 15h3\"\/><\/svg><\/div>\n            <h3 style=\"font-size:15px;margin-bottom:4px;\">Schneller<\/h3>\n            <p style=\"color:var(--invert-soft);font-size:13.5px;\">als general-purpose LLMs<\/p>\n          <\/div>\n          <div class=\"deploy-card\" style=\"padding:20px;\">\n            <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"9\" cy=\"9\" r=\"5.5\"\/><circle cx=\"15\" cy=\"15\" r=\"5.5\"\/><\/svg><\/div>\n            <h3 style=\"font-size:15px;margin-bottom:4px;\">G\u00fcnstiger<\/h3>\n            <p style=\"color:var(--invert-soft);font-size:13.5px;\">im Betrieb<\/p>\n          <\/div>\n          <div class=\"deploy-card\" style=\"padding:20px;\">\n            <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"5\" y=\"10.5\" width=\"14\" height=\"10\" rx=\"1.5\"\/><path d=\"M8 10.5V7a4 4 0 1 1 8 0v3.5\"\/><\/svg><\/div>\n            <h3 style=\"font-size:15px;margin-bottom:4px;\">Air-Gapped<\/h3>\n            <p style=\"color:var(--invert-soft);font-size:13.5px;\">auf 1 GPU (24 GB)<\/p>\n          <\/div>\n        <\/div>\n      <\/div>\n    <\/div>\n    <div class=\"linkcards\">\n      <a class=\"linkcard\" href=\"#benchmarks\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M9 12.5l2 2 4-4.5\"\/><circle cx=\"12\" cy=\"12\" r=\"9\"\/><\/svg><\/div>\n        <div><h4>Benchmark Report lesen<\/h4><p>Vollst\u00e4ndige Ergebnisse je Branche und Modell, Konfidenzintervalle und Methodik.<\/p><span class=\"go\">Anfragen \u2192<\/span><\/div>\n      <\/a>\n      <a class=\"linkcard\" href=\"https:\/\/playground.parashift.io\/\" target=\"_blank\" rel=\"noopener\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><rect x=\"6\" y=\"6\" width=\"12\" height=\"12\" rx=\"2\"\/><path d=\"M9 3v3M15 3v3M9 18v3M15 18v3M3 9h3M3 15h3M18 9h3M18 15h3\"\/><\/svg><\/div>\n        <div><h4>Modell testen<\/h4><p>Eigene Dokumente ausprobieren unter playground.parashift.io.<\/p><span class=\"go\">Playground \u00f6ffnen \u2192<\/span><\/div>\n      <\/a>\n      <a class=\"linkcard\" href=\"https:\/\/dokumentli-docs.parashift.io\/\" target=\"_blank\" rel=\"noopener\">\n        <div class=\"icon\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M7 3h7l4 4v13a1 1 0 0 1-1 1H7a1 1 0 0 1-1-1V4a1 1 0 0 1 1-1Z\"\/><path d=\"M14 3v4h4\"\/><path d=\"M9 12h6\"\/><path d=\"M9 15.5h6\"\/><\/svg><\/div>\n        <div><h4>Dokumentation<\/h4><p>Technische Details unter dokumentli-docs.parashift.io.<\/p><span class=\"go\">Docs \u00f6ffnen \u2192<\/span><\/div>\n      <\/a>\n    <\/div>\n  <\/div>\n<\/section>\n\n<section class=\"alt\">\n  <div class=\"wrap\">\n    <div class=\"section-head\">\n      <span class=\"eyebrow\">H\u00e4ufige Fragen<\/span>\n      <h2>Was Automatisierungsverantwortliche uns zuerst fragen<\/h2>\n    <\/div>\n    <div class=\"faq\">\n      <div class=\"faqitem open\">\n        <button class=\"q\" type=\"button\">\n          <span>Ist Dokumentli\u00ae wirklich vollst\u00e4ndig air-gapped nutzbar?<\/span>\n          <span class=\"plus\">+<\/span>\n        <\/button>\n        <div class=\"a\"><p>Ja. In der Einsatzoption On-Prem \/ eigene Hardware l\u00e4uft Dokumentli\u00ae im eigenen Rechenzentrum \u00fcber den Dokumentli\u00ae Node, keine Daten verlassen das eigene Netzwerk, und eine einzelne konsumenten-taugliche GPU mit 24 GB VRAM reicht aus. Es ist weder ein Rechenzentrum noch eine Internetanbindung n\u00f6tig.<\/p><\/div>\n      <\/div>\n      <div class=\"faqitem\">\n        <button class=\"q\" type=\"button\">\n          <span>Wie wurde die Genauigkeit von 88,4% gemessen?<\/span>\n          <span class=\"plus\">+<\/span>\n        <\/button>\n        <div class=\"a\"><p>In einem unabh\u00e4ngigen Test auf 5 Benchmark-Datens\u00e4tzen mit insgesamt 3'368 realen Gesch\u00e4ftsdokumenten, gegen sieben Modelle inklusive Dokumentli\u00ae, mit cANLS@0.8 (Average Normalized Levenshtein Similarity mit Schwellenwert 0.8) als G\u00fctema\u00df. Die vollst\u00e4ndige Methodik und Detailergebnisse sind im separaten Benchmark Report dokumentiert.<\/p><\/div>\n      <\/div>\n      <div class=\"faqitem\">\n        <button class=\"q\" type=\"button\">\n          <span>Kann Dokumentli\u00ae mit unserem bestehenden LLM-Setup kombiniert werden?<\/span>\n          <span class=\"plus\">+<\/span>\n        <\/button>\n        <div class=\"a\"><p>Ja, kombinierbar statt exklusiv. Viele Unternehmen setzen Dokumentli\u00ae gezielt f\u00fcr die Dokumenteneingangsverarbeitung ein und kombinieren es bei Bedarf mit general-purpose Modellen f\u00fcr nachgelagerte, dokumentfremde Schritte wie offenes Reasoning oder Code.<\/p><\/div>\n      <\/div>\n      <div class=\"faqitem\">\n        <button class=\"q\" type=\"button\">\n          <span>Wie wird Dokumentli\u00ae abgerechnet?<\/span>\n          <span class=\"plus\">+<\/span>\n        <\/button>\n        <div class=\"a\"><p>\u00dcber ein Abonnement mit Seitenkontingent statt Pay-per-Call, gestaffelt nach monatlichem Dokumentenvolumen von Pilotprojekten bis zum Hochvolumen-Einsatz. Jede Stufe umfasst 4 Basismodell-Updates pro Jahr, ohne neuen Vertrag.<\/p><\/div>\n      <\/div>\n      <div class=\"faqitem\">\n        <button class=\"q\" type=\"button\">\n          <span>Welche Einsatzoptionen gibt es?<\/span>\n          <span class=\"plus\">+<\/span>\n        <\/button>\n        <div class=\"a\"><p>Vier: On-Prem \/ eigene Hardware und Private-Cloud-Instanz sind heute verf\u00fcgbar, die Parashift Plattform und die Dokumentli\u00ae API folgen demn\u00e4chst. Alle vier lassen sich mit praktisch jeder Dokumenten- oder Workflow-Automatisierungsl\u00f6sung verbinden.<\/p><\/div>\n      <\/div>\n      <div class=\"faqitem\">\n        <button class=\"q\" type=\"button\">\n          <span>Was kann Dokumentli\u00ae nicht?<\/span>\n          <span class=\"plus\">+<\/span>\n        <\/button>\n        <div class=\"a\"><p>Dokumentli\u00ae ist kein general-purpose Assistent: F\u00fcr freies Reasoning, Code-Generierung, allgemeine Wissensfragen, freies kreatives Schreiben oder mehrstufige dokumentfremde Agenten-Aufgaben ist ein general-purpose Modell die richtige Wahl.<\/p><\/div>\n      <\/div>\n    <\/div>\n  <\/div>\n<\/section>\n<script type=\"application\/ld+json\">\n{\n  \"@context\": \"https:\/\/schema.org\",\n  \"@type\": \"FAQPage\",\n  \"mainEntity\": [\n    {\n      \"@type\": \"Question\",\n      \"name\": \"Ist Dokumentli\u00ae wirklich vollst\u00e4ndig air-gapped nutzbar?\",\n      \"acceptedAnswer\": {\n        \"@type\": \"Answer\",\n        \"text\": \"Ja. In der Einsatzoption On-Prem \/ eigene Hardware l\u00e4uft Dokumentli\u00ae im eigenen Rechenzentrum \u00fcber den Dokumentli\u00ae Node, keine Daten verlassen das eigene Netzwerk, und eine einzelne konsumenten-taugliche GPU mit 24 GB VRAM reicht aus.\"\n      }\n    },\n    {\n      \"@type\": \"Question\",\n      \"name\": \"Wie wurde die Genauigkeit von 88,4% gemessen?\",\n      \"acceptedAnswer\": {\n        \"@type\": \"Answer\",\n        \"text\": \"In einem unabh\u00e4ngigen Test auf 5 Benchmark-Datens\u00e4tzen mit insgesamt 3'368 realen Gesch\u00e4ftsdokumenten, gegen sieben Modelle inklusive Dokumentli\u00ae, mit cANLS@0.8 als G\u00fctema\u00df.\"\n      }\n    },\n    {\n      \"@type\": \"Question\",\n      \"name\": \"Kann Dokumentli\u00ae mit unserem bestehenden LLM-Setup kombiniert werden?\",\n      \"acceptedAnswer\": {\n        \"@type\": \"Answer\",\n        \"text\": \"Ja, kombinierbar statt exklusiv. Dokumentli\u00ae wird f\u00fcr die Dokumenteneingangsverarbeitung eingesetzt und l\u00e4sst sich mit general-purpose Modellen f\u00fcr nachgelagerte, dokumentfremde Schritte kombinieren.\"\n      }\n    },\n    {\n      \"@type\": \"Question\",\n      \"name\": \"Wie wird Dokumentli\u00ae abgerechnet?\",\n      \"acceptedAnswer\": {\n        \"@type\": \"Answer\",\n        \"text\": \"\u00dcber ein Abonnement mit Seitenkontingent statt Pay-per-Call, gestaffelt nach monatlichem Dokumentenvolumen, inklusive 4 Basismodell-Updates pro Jahr.\"\n      }\n    },\n    {\n      \"@type\": \"Question\",\n      \"name\": \"Welche Einsatzoptionen gibt es?\",\n      \"acceptedAnswer\": {\n        \"@type\": \"Answer\",\n        \"text\": \"Vier: On-Prem \/ eigene Hardware und Private-Cloud-Instanz sind heute verf\u00fcgbar, die Parashift Plattform und die Dokumentli API folgen demn\u00e4chst.\"\n      }\n    },\n    {\n      \"@type\": \"Question\",\n      \"name\": \"Was kann Dokumentli\u00ae nicht?\",\n      \"acceptedAnswer\": {\n        \"@type\": \"Answer\",\n        \"text\": \"Dokumentli ist kein general-purpose Assistent: f\u00fcr freies Reasoning, Code-Generierung, allgemeine Wissensfragen, freies kreatives Schreiben oder dokumentfremde Agenten-Aufgaben ist ein general-purpose Modell die richtige Wahl.\"\n      }\n    }\n  ]\n}\n<\/script>\n\n<section>\n  <div class=\"wrap\">\n    <div class=\"finalcta\">\n      <h2>Testen Sie Dokumentli\u00ae an Ihren eigenen Dokumenten.<\/h2>\n      <p>30 Sekunden im Playground zeigen mehr als jede Folie: Laden Sie ein eigenes Dokument hoch und sehen Sie, wie Dokumentli\u00ae es liest, klassifiziert und strukturiert.<\/p>\n      <div class=\"ctas\">\n        <a class=\"btn\" href=\"https:\/\/playground.parashift.io\/\" target=\"_blank\" rel=\"noopener\">Modell im Playground testen<\/a>\n        <a class=\"btn ghost\" href=\"\/de\/demo\/\">30-Min Demo buchen<\/a>\n      <\/div>\n      <div><span class=\"cite\">Benchmark Report auf Anfrage \u00b7 Dokumentation unter dokumentli-docs.parashift.io<\/span><\/div>\n    <\/div>\n  <\/div>\n<\/section>\n\n<\/div>\n\n<script>\n(function(){\n  var items = document.querySelectorAll('.psvr-scope .faqitem .q');\n  for (var i = 0; i < items.length; i++){\n    items[i].addEventListener('click', function(){\n      this.closest('.faqitem').classList.toggle('open');\n    });\n  }\n  if (window.matchMedia && matchMedia('(prefers-reduced-motion: reduce)').matches){\n    document.querySelectorAll('.psvr-scope .mv-svg').forEach(function(svg){\n      if (svg.pauseAnimations) svg.pauseAnimations();\n    });\n  }\n})();\n<\/script>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<\/div>\n\t\t","protected":false},"excerpt":{"rendered":"<p>Dokumentli\u00ae \u00b7 Sovereign Document AI Dokumentli\u00ae \u2013 Das KI-Modell, das LLMs bei Dokumenten schl\u00e4gt. Dokumentli\u00ae liest, klassifiziert und strukturiert Gesch\u00e4ftsdokumente pr\u00e4ziser und schneller als general-purpose Frontier-Modelle \u2013 und l\u00e4uft wahlweise vollst\u00e4ndig air-gapped auf einer einzelnen Consumer-GPU mit 24 GB VRAM&#8230;.<\/p>\n","protected":false},"author":1,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"content-type":"","footnotes":""},"class_list":["post-50052","page","type-page","status-publish","hentry"],"_links":{"self":[{"href":"https:\/\/parashift.ai\/de\/wp-json\/wp\/v2\/pages\/50052","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/parashift.ai\/de\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/parashift.ai\/de\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/parashift.ai\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/parashift.ai\/de\/wp-json\/wp\/v2\/comments?post=50052"}],"version-history":[{"count":21,"href":"https:\/\/parashift.ai\/de\/wp-json\/wp\/v2\/pages\/50052\/revisions"}],"predecessor-version":[{"id":51571,"href":"https:\/\/parashift.ai\/de\/wp-json\/wp\/v2\/pages\/50052\/revisions\/51571"}],"wp:attachment":[{"href":"https:\/\/parashift.ai\/de\/wp-json\/wp\/v2\/media?parent=50052"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}