Files
engine/site/index.html
zeekayandHanzo Dev 51b1849ba1 docs: every throughput figure renders from the run that produced it
Six claim pages, one per measurement, each fetching its own rows from
/v1/benchmark/throughput at read time and printing the median with its
range, the repetition count, and the engine build, flags, host and date
that stand behind it. A row whose quality checks failed renders
"withheld" in place of a rate, so a fast run that was serving badly
cannot be quoted.

Drops the figures that had no run behind them: the llama.cpp comparison
table in the ROCm guide, the tok/s cards and the "2-4x" feature stats on
the marketing page, the backend table in gpu/README, and MATFORMER's
layer-count speed estimate.

Claims are reached from the footer of every page rather than the sidebar,
so one is always a hop away and never quoted inline.

Co-authored-by: Hanzo Dev <[email protected]>
2026-09-16 02:51:11 -07:00

1107 lines
37 KiB
HTML

<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Hanzo Engine — Production AI Inference at Any Scale</title>
<meta name="description" content="High-performance production inference serving Zen models and 60+ architectures. CUDA, Metal, and CPU backends. OpenAI-compatible API. Deploy anywhere.">
<meta name="keywords" content="AI inference, LLM serving, CUDA, production AI, Zen models, OpenAI compatible, model serving, GPU inference">
<meta property="og:title" content="Hanzo Engine — Production AI Inference at Any Scale">
<meta property="og:description" content="Serve Zen models and 60+ architectures with CUDA, Metal, and CPU backends. OpenAI-compatible API.">
<meta property="og:type" content="website">
<meta property="og:url" content="https://engine.hanzo.ai">
<link rel="icon" href="data:image/svg+xml,<svg viewBox='0 0 64 64' xmlns='http://www.w3.org/2000/svg'><rect width='64' height='64' rx='8' fill='%23000'/><g transform='translate(8,8) scale(0.716)'><path d='M22.21 67V44.6369H0V67H22.21Z' fill='%23fff'/><path d='M66.7038 22.3184H22.2534L0.0878906 44.6367H44.4634L66.7038 22.3184Z' fill='%23fff'/><path d='M22.21 0H0V22.3184H22.21V0Z' fill='%23fff'/><path d='M66.7198 0H44.5098V22.3184H66.7198V0Z' fill='%23fff'/><path d='M66.7198 67V44.6369H44.5098V67H66.7198Z' fill='%23fff'/></g></svg>">
<style>
*, *::before, *::after { margin: 0; padding: 0; box-sizing: border-box; }
body {
font-family: Zen, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif;
background: #000;
color: #f5f5f5;
line-height: 1.6;
-webkit-font-smoothing: antialiased;
}
a { color: #f5f5f5; text-decoration: none; }
a:hover { text-decoration: underline; }
code, .mono {
font-family: Zen Mono, ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", monospace;
}
.container {
max-width: 1060px;
margin: 0 auto;
padding: 0 24px;
}
/* ---- Nav ---- */
nav {
border-bottom: 1px solid #1a1a1a;
padding: 16px 0;
position: sticky;
top: 0;
background: rgba(0,0,0,0.92);
backdrop-filter: blur(12px);
-webkit-backdrop-filter: blur(12px);
z-index: 100;
}
nav .container {
display: flex;
justify-content: space-between;
align-items: center;
}
.nav-brand {
font-size: 15px;
font-weight: 600;
color: #fff;
letter-spacing: -0.02em;
}
.nav-links {
display: flex;
gap: 24px;
list-style: none;
}
.nav-links a {
font-size: 13px;
color: #888;
transition: color 0.15s;
}
.nav-links a:hover { color: #fff; text-decoration: none; }
/* ---- Hero ---- */
.hero {
padding: 100px 0 88px;
text-align: center;
border-bottom: 1px solid #1a1a1a;
}
.hero-badge {
display: inline-block;
font-size: 11px;
font-weight: 600;
text-transform: uppercase;
letter-spacing: 0.12em;
color: #888;
border: 1px solid #333;
border-radius: 100px;
padding: 6px 16px;
margin-bottom: 28px;
}
.hero h1 {
font-size: 56px;
font-weight: 700;
color: #fff;
letter-spacing: -0.035em;
margin-bottom: 16px;
line-height: 1.1;
}
.hero .tagline {
font-size: 20px;
color: #888;
max-width: 620px;
margin: 0 auto 20px;
line-height: 1.5;
}
.hero .sub-tagline {
font-size: 14px;
color: #555;
max-width: 560px;
margin: 0 auto 44px;
}
.hero-install {
display: inline-block;
background: #111;
border: 1px solid #333;
border-radius: 8px;
padding: 14px 24px;
font-family: Zen Mono, ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", monospace;
font-size: 14px;
color: #f5f5f5;
cursor: pointer;
transition: border-color 0.15s;
position: relative;
margin-bottom: 28px;
}
.hero-install:hover { border-color: #555; text-decoration: none; }
.hero-install .copy-hint {
font-family: Zen, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif;
font-size: 11px;
color: #555;
margin-left: 16px;
}
.hero-ctas {
display: flex;
justify-content: center;
gap: 16px;
flex-wrap: wrap;
}
.btn {
display: inline-flex;
align-items: center;
gap: 8px;
padding: 12px 28px;
border-radius: 8px;
font-size: 14px;
font-weight: 500;
transition: all 0.15s;
cursor: pointer;
border: none;
}
.btn-primary {
background: #fff;
color: #000;
}
.btn-primary:hover { background: #ddd; text-decoration: none; color: #000; }
.btn-secondary {
background: transparent;
color: #f5f5f5;
border: 1px solid #333;
}
.btn-secondary:hover { border-color: #555; text-decoration: none; }
/* ---- Sections ---- */
section {
padding: 80px 0;
border-bottom: 1px solid #1a1a1a;
}
section:last-of-type { border-bottom: none; }
.section-label {
font-size: 11px;
font-weight: 600;
text-transform: uppercase;
letter-spacing: 0.1em;
color: #555;
margin-bottom: 12px;
}
.section-title {
font-size: 32px;
font-weight: 700;
color: #fff;
letter-spacing: -0.025em;
margin-bottom: 16px;
line-height: 1.2;
}
.section-desc {
font-size: 15px;
color: #888;
max-width: 640px;
margin-bottom: 48px;
line-height: 1.6;
}
/* ---- Value Props (4-col) ---- */
.value-grid {
display: grid;
grid-template-columns: repeat(4, 1fr);
gap: 16px;
}
.value-card {
background: #0a0a0a;
border: 1px solid #1a1a1a;
border-radius: 8px;
padding: 28px 24px;
transition: border-color 0.15s;
}
.value-card:hover { border-color: #333; }
.value-icon {
font-size: 24px;
margin-bottom: 16px;
display: block;
color: #888;
}
.value-card h3 {
font-size: 15px;
font-weight: 600;
color: #fff;
margin-bottom: 8px;
}
.value-card p {
font-size: 13px;
color: #888;
line-height: 1.55;
}
/* ---- Feature Grid (3-col) ---- */
.feature-grid {
display: grid;
grid-template-columns: repeat(3, 1fr);
gap: 16px;
}
.feature-card {
background: #0a0a0a;
border: 1px solid #1a1a1a;
border-radius: 8px;
padding: 28px 24px;
transition: border-color 0.15s;
}
.feature-card:hover { border-color: #333; }
.feature-card h3 {
font-size: 15px;
font-weight: 600;
color: #fff;
margin-bottom: 8px;
}
.feature-card .feat-stat {
font-size: 12px;
color: #555;
margin-bottom: 12px;
font-weight: 500;
}
.feature-card p {
font-size: 13px;
color: #888;
line-height: 1.55;
}
/* ---- Zen Models ---- */
.model-grid {
display: grid;
grid-template-columns: repeat(3, 1fr);
gap: 16px;
}
.model-card {
background: #0a0a0a;
border: 1px solid #1a1a1a;
border-radius: 8px;
padding: 24px;
transition: border-color 0.15s;
}
.model-card:hover { border-color: #333; }
.model-card h3 {
font-size: 16px;
font-weight: 600;
color: #fff;
margin-bottom: 4px;
}
.model-card .model-desc {
font-size: 12px;
color: #888;
margin-bottom: 16px;
}
.model-meta {
display: flex;
gap: 16px;
margin-bottom: 16px;
}
.model-meta span {
font-size: 11px;
color: #555;
font-weight: 500;
}
.model-meta .meta-val {
color: #ccc;
font-family: Zen Mono, ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", monospace;
}
.model-snippet {
background: #111;
border: 1px solid #1a1a1a;
border-radius: 6px;
padding: 12px 14px;
overflow-x: auto;
}
.model-snippet pre {
font-family: Zen Mono, ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", monospace;
font-size: 11px;
line-height: 1.5;
color: #888;
}
.model-snippet .cmd { color: #ccc; }
.models-footer {
text-align: center;
margin-top: 32px;
}
.models-footer a {
font-size: 13px;
color: #888;
border: 1px solid #333;
border-radius: 8px;
padding: 10px 24px;
display: inline-block;
transition: all 0.15s;
}
.models-footer a:hover { color: #fff; border-color: #555; text-decoration: none; }
/* ---- Deploy Tabs ---- */
.deploy-tabs {
display: flex;
gap: 0;
border-bottom: 1px solid #1a1a1a;
margin-bottom: 0;
}
.deploy-tab {
padding: 10px 20px;
font-size: 13px;
font-weight: 500;
color: #555;
cursor: pointer;
border-bottom: 2px solid transparent;
transition: all 0.15s;
background: none;
border-top: none;
border-left: none;
border-right: none;
font-family: Zen, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif;
}
.deploy-tab:hover { color: #888; }
.deploy-tab.active {
color: #fff;
border-bottom-color: #fff;
}
.deploy-panel {
display: none;
}
.deploy-panel.active {
display: block;
}
/* ---- Code Blocks ---- */
.code-block {
background: #111;
border: 1px solid #1a1a1a;
border-radius: 0 0 8px 8px;
padding: 24px 28px;
overflow-x: auto;
}
.code-block.standalone {
border-radius: 8px;
}
.code-block pre {
font-family: Zen Mono, ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", monospace;
font-size: 13px;
line-height: 1.7;
color: #ccc;
}
.code-block .comment { color: #555; }
.code-block .cmd { color: #f5f5f5; }
.code-block .str { color: #aaa; }
.code-block .key { color: #888; }
.code-label {
font-size: 11px;
font-weight: 600;
text-transform: uppercase;
letter-spacing: 0.08em;
color: #555;
margin-bottom: 8px;
}
/* ---- Tables ---- */
.table-wrap {
overflow-x: auto;
border: 1px solid #1a1a1a;
border-radius: 8px;
}
table {
width: 100%;
border-collapse: collapse;
font-size: 13px;
}
thead th {
text-align: left;
padding: 12px 16px;
font-weight: 600;
font-size: 12px;
text-transform: uppercase;
letter-spacing: 0.06em;
color: #555;
background: #0a0a0a;
border-bottom: 1px solid #1a1a1a;
}
tbody td {
padding: 10px 16px;
color: #888;
border-bottom: 1px solid #0f0f0f;
}
tbody tr:last-child td { border-bottom: none; }
tbody tr:hover { background: #0a0a0a; }
td:first-child { color: #fff; font-weight: 500; }
td code {
font-size: 12px;
background: #111;
padding: 2px 6px;
border-radius: 4px;
}
/* ---- API Reference ---- */
.endpoint-list {
display: flex;
flex-direction: column;
gap: 8px;
margin-bottom: 32px;
}
.endpoint {
display: flex;
align-items: center;
gap: 12px;
background: #0a0a0a;
border: 1px solid #1a1a1a;
border-radius: 8px;
padding: 12px 20px;
transition: border-color 0.15s;
}
.endpoint:hover { border-color: #333; }
.endpoint .method {
font-family: Zen Mono, ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", monospace;
font-size: 11px;
font-weight: 700;
color: #888;
min-width: 44px;
}
.endpoint .path {
font-family: Zen Mono, ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", monospace;
font-size: 13px;
color: #f5f5f5;
}
.endpoint .ep-desc {
font-size: 12px;
color: #555;
margin-left: auto;
}
.api-examples {
display: grid;
grid-template-columns: repeat(2, 1fr);
gap: 16px;
}
.api-examples .code-block {
border-radius: 8px;
}
.api-example-label {
font-size: 11px;
font-weight: 600;
text-transform: uppercase;
letter-spacing: 0.08em;
color: #555;
margin-bottom: 8px;
}
/* ---- Comparison ---- */
.comparison-table thead th:first-child { width: 180px; }
.comparison-table thead th:not(:first-child) { text-align: center; }
.comparison-table tbody td:not(:first-child) { text-align: center; color: #ccc; }
/* ---- Footer ---- */
footer {
border-top: 1px solid #1a1a1a;
padding: 40px 0;
}
footer .container {
display: flex;
justify-content: space-between;
align-items: center;
flex-wrap: wrap;
gap: 16px;
}
.footer-copy {
font-size: 12px;
color: #555;
}
.footer-links {
display: flex;
gap: 24px;
list-style: none;
}
.footer-links a {
font-size: 12px;
color: #666;
transition: color 0.15s;
}
.footer-links a:hover { color: #fff; text-decoration: none; }
/* ---- Responsive ---- */
@media (max-width: 900px) {
.value-grid { grid-template-columns: repeat(2, 1fr); }
.model-grid { grid-template-columns: repeat(2, 1fr); }
.api-examples { grid-template-columns: 1fr; }
}
@media (max-width: 768px) {
.hero { padding: 72px 0 64px; }
.hero h1 { font-size: 36px; }
.hero .tagline { font-size: 17px; }
.feature-grid { grid-template-columns: 1fr; }
.model-grid { grid-template-columns: 1fr; }
section { padding: 56px 0; }
.section-title { font-size: 26px; }
.nav-links { gap: 16px; }
}
@media (max-width: 480px) {
.hero h1 { font-size: 30px; }
.hero-install { font-size: 11px; padding: 12px 14px; }
.hero-install .copy-hint { display: none; }
.nav-links { display: none; }
.value-grid { grid-template-columns: 1fr; }
footer .container { flex-direction: column; align-items: flex-start; }
.hero-ctas { flex-direction: column; align-items: center; }
.endpoint .ep-desc { display: none; }
}
</style>
</head>
<body>
<!-- Nav -->
<nav>
<div class="container">
<a class="nav-brand" href="/" style="display:flex;align-items:center;gap:8px"><svg viewBox="0 0 67 67" style="width:20px;height:20px" xmlns="http://www.w3.org/2000/svg"><path d="M22.21 67V44.6369H0V67H22.21Z" fill="#fff"/><path d="M0 44.6369L22.21 46.8285V44.6369H0Z" fill="#DDD"/><path d="M66.7038 22.3184H22.2534L0.0878906 44.6367H44.4634L66.7038 22.3184Z" fill="#fff"/><path d="M22.21 0H0V22.3184H22.21V0Z" fill="#fff"/><path d="M66.7198 0H44.5098V22.3184H66.7198V0Z" fill="#fff"/><path d="M66.6753 22.3185L44.5098 20.0822V22.3185H66.6753Z" fill="#DDD"/><path d="M66.7198 67V44.6369H44.5098V67H66.7198Z" fill="#fff"/></svg>Hanzo Engine</a>
<ul class="nav-links">
<li><a href="#features">Features</a></li>
<li><a href="#models">Models</a></li>
<li><a href="#deploy">Deploy</a></li>
<li><a href="#architecture">Architecture</a></li>
<li><a href="#api">API</a></li>
<li><a href="https://github.com/hanzoai/engine">GitHub</a></li>
</ul>
</div>
</nav>
<!-- ============================================================ -->
<!-- 1. Hero -->
<!-- ============================================================ -->
<div class="hero">
<div class="container">
<span class="hero-badge">Cloud Inference Runtime</span>
<h1>Hanzo Engine</h1>
<p class="tagline">Production AI inference at any scale</p>
<p class="sub-tagline">Serve Zen models and 60+ architectures with CUDA, Metal, and CPU backends. OpenAI-compatible API. Deploy anywhere.</p>
<a class="hero-install" id="installCmd" onclick="copyInstall(event)">
curl -sSL https://engine.hanzo.ai/install.sh | sh
<span class="copy-hint">click to copy</span>
</a>
<div class="hero-ctas">
<a class="btn btn-primary" href="#deploy">Get Started</a>
<a class="btn btn-secondary" href="https://github.com/hanzoai/engine/pkgs/container/engine">Docker Hub</a>
</div>
</div>
</div>
<!-- ============================================================ -->
<!-- 2. Value Props (4 columns) -->
<!-- ============================================================ -->
<section id="features">
<div class="container">
<div class="section-label">Why Engine</div>
<h2 class="section-title">Built for production inference</h2>
<p class="section-desc">Everything you need to serve large language models at scale, with zero configuration overhead.</p>
<div class="value-grid">
<div class="value-card">
<span class="value-icon">&#9633;</span>
<h3>60+ Architectures</h3>
<p>Zen, Llama, Qwen3, Mistral, DeepSeek, Gemma, Phi — all major model families. Load from HuggingFace or local files.</p>
</div>
<div class="value-card">
<span class="value-icon">&#9653;</span>
<h3>GPU Accelerated</h3>
<p>CUDA 12+, Metal, CPU with AVX2/AVX-512. Automatic backend selection based on detected hardware.</p>
</div>
<div class="value-card">
<span class="value-icon">&#9671;</span>
<h3>Production Ready</h3>
<p>Paged attention v2, continuous batching, speculative decoding, tensor parallelism. Built for throughput.</p>
</div>
<div class="value-card">
<span class="value-icon">&#9675;</span>
<h3>OpenAI Compatible</h3>
<p>Drop-in replacement. Same API, same client libraries, your infrastructure. No code changes required.</p>
</div>
</div>
</div>
</section>
<!-- ============================================================ -->
<!-- 3. Zen Models -- Optimized for Engine -->
<!-- ============================================================ -->
<section id="models">
<div class="container">
<div class="section-label">Zen Models</div>
<h2 class="section-title">Optimized for Engine</h2>
<p class="section-desc">First-class support for every Zen model. Pre-optimized serving profiles, automatic quantization selection, and native architecture support.</p>
<div class="model-grid">
<div class="model-card">
<h3>zen4</h3>
<div class="model-desc">Flagship reasoning</div>
<div class="model-meta">
<span>Params <span class="meta-val">744B MoE</span></span>
<span>Context <span class="meta-val">202K</span></span>
</div>
<div class="model-snippet">
<pre><span class="cmd">hanzo-engine serve \
--model zenlm/zen4 \
--port 8000</span></pre>
</div>
</div>
<div class="model-card">
<h3>zen4-max</h3>
<div class="model-desc">Maximum capability</div>
<div class="model-meta">
<span>Params <span class="meta-val">1.04T MoE</span></span>
<span>Context <span class="meta-val">256K</span></span>
</div>
<div class="model-snippet">
<pre><span class="cmd">hanzo-engine serve \
--model zenlm/zen4-max \
--port 8000</span></pre>
</div>
</div>
<div class="model-card">
<h3>zen4-coder</h3>
<div class="model-desc">Code generation</div>
<div class="model-meta">
<span>Params <span class="meta-val">480B MoE</span></span>
<span>Context <span class="meta-val">262K</span></span>
</div>
<div class="model-snippet">
<pre><span class="cmd">hanzo-engine serve \
--model zenlm/zen4-coder \
--port 8000</span></pre>
</div>
</div>
<div class="model-card">
<h3>zen4-mini</h3>
<div class="model-desc">Fast and efficient</div>
<div class="model-meta">
<span>Params <span class="meta-val">8B dense</span></span>
<span>Context <span class="meta-val">40K</span></span>
</div>
<div class="model-snippet">
<pre><span class="cmd">hanzo-engine serve \
--model zenlm/zen4-mini \
--port 8000</span></pre>
</div>
</div>
<div class="model-card">
<h3>zen4-ultra</h3>
<div class="model-desc">Extended thinking</div>
<div class="model-meta">
<span>Params <span class="meta-val">744B MoE + CoT</span></span>
<span>Context <span class="meta-val">202K</span></span>
</div>
<div class="model-snippet">
<pre><span class="cmd">hanzo-engine serve \
--model zenlm/zen4-ultra \
--port 8000</span></pre>
</div>
</div>
<div class="model-card">
<h3>zen3-omni</h3>
<div class="model-desc">Multimodal</div>
<div class="model-meta">
<span>Params <span class="meta-val">~200B</span></span>
<span>Context <span class="meta-val">202K</span></span>
</div>
<div class="model-snippet">
<pre><span class="cmd">hanzo-engine serve \
--model zenlm/zen3-omni \
--port 8000</span></pre>
</div>
</div>
</div>
<div class="models-footer">
<a href="https://huggingface.co/zenlm">Browse all models on HuggingFace &rarr;</a>
</div>
</div>
</section>
<!-- ============================================================ -->
<!-- 4. Deployment Options (tabbed) -->
<!-- ============================================================ -->
<section id="deploy">
<div class="container">
<div class="section-label">Deployment</div>
<h2 class="section-title">Deploy anywhere</h2>
<p class="section-desc">From a single CLI command to a multi-replica Kubernetes deployment. Choose the method that fits your infrastructure.</p>
<div class="deploy-tabs">
<button class="deploy-tab active" onclick="switchTab(event, 'tab-cli')">CLI</button>
<button class="deploy-tab" onclick="switchTab(event, 'tab-docker')">Docker</button>
<button class="deploy-tab" onclick="switchTab(event, 'tab-k8s')">Kubernetes</button>
</div>
<div class="deploy-panel active" id="tab-cli">
<div class="code-block">
<pre><span class="comment"># Install</span>
<span class="cmd">curl -sSL https://engine.hanzo.ai/install.sh | sh</span>
<span class="comment"># Or via cargo</span>
<span class="cmd">cargo install hanzo-engine</span>
<span class="comment"># Serve a model</span>
<span class="cmd">hanzo-engine serve --model zenlm/zen4-mini --port 8000</span>
<span class="comment"># With CUDA backend</span>
<span class="cmd">hanzo-engine serve --model zenlm/zen4 --port 8000 --features cuda</span>
<span class="comment"># With tensor parallelism across 4 GPUs</span>
<span class="cmd">hanzo-engine serve --model zenlm/zen4 --port 8000 --tp 4</span></pre>
</div>
</div>
<div class="deploy-panel" id="tab-docker">
<div class="code-block">
<pre><span class="comment"># Pull and run with GPU support</span>
<span class="cmd">docker run -p 8000:8000 --gpus all \
ghcr.io/hanzoai/engine:latest \
serve --model zenlm/zen4 --port 8000</span>
<span class="comment"># With volume mount for local models</span>
<span class="cmd">docker run -p 8000:8000 --gpus all \
-v /models:/models \
ghcr.io/hanzoai/engine:latest \
serve --model /models/zen4-mini --port 8000</span>
<span class="comment"># CPU-only</span>
<span class="cmd">docker run -p 8000:8000 \
ghcr.io/hanzoai/engine:latest \
serve --model zenlm/zen4-mini --port 8000</span></pre>
</div>
</div>
<div class="deploy-panel" id="tab-k8s">
<div class="code-block">
<pre><span class="key">apiVersion:</span> <span class="str">apps/v1</span>
<span class="key">kind:</span> <span class="str">Deployment</span>
<span class="key">metadata:</span>
<span class="key">name:</span> <span class="str">hanzo-engine</span>
<span class="key">spec:</span>
<span class="key">replicas:</span> <span class="str">2</span>
<span class="key">selector:</span>
<span class="key">matchLabels:</span>
<span class="key">app:</span> <span class="str">hanzo-engine</span>
<span class="key">template:</span>
<span class="key">metadata:</span>
<span class="key">labels:</span>
<span class="key">app:</span> <span class="str">hanzo-engine</span>
<span class="key">spec:</span>
<span class="key">containers:</span>
- <span class="key">name:</span> <span class="str">engine</span>
<span class="key">image:</span> <span class="str">ghcr.io/hanzoai/engine:latest</span>
<span class="key">args:</span> <span class="str">["serve", "--model", "zenlm/zen4", "--port", "8000"]</span>
<span class="key">ports:</span>
- <span class="key">containerPort:</span> <span class="str">8000</span>
<span class="key">resources:</span>
<span class="key">limits:</span>
<span class="key">nvidia.com/gpu:</span> <span class="str">1</span>
---
<span class="key">apiVersion:</span> <span class="str">v1</span>
<span class="key">kind:</span> <span class="str">Service</span>
<span class="key">metadata:</span>
<span class="key">name:</span> <span class="str">hanzo-engine</span>
<span class="key">spec:</span>
<span class="key">selector:</span>
<span class="key">app:</span> <span class="str">hanzo-engine</span>
<span class="key">ports:</span>
- <span class="key">port:</span> <span class="str">8000</span>
<span class="key">targetPort:</span> <span class="str">8000</span></pre>
</div>
</div>
</div>
</section>
<!-- ============================================================ -->
<!-- 5. Architecture Deep Dive -->
<!-- ============================================================ -->
<section id="architecture">
<div class="container">
<div class="section-label">Internals</div>
<h2 class="section-title">Architecture deep dive</h2>
<p class="section-desc">Purpose-built inference primitives that maximize throughput and minimize latency at every layer of the stack.</p>
<div class="feature-grid">
<div class="feature-card">
<h3>Paged Attention v2</h3>
<div class="feat-stat">Non-contiguous KV blocks</div>
<p>Efficient KV-cache management with non-contiguous memory blocks. Serves more concurrent requests with less GPU memory. Eliminates fragmentation waste.</p>
</div>
<div class="feature-card">
<h3>Continuous Batching</h3>
<div class="feat-stat">Maximum GPU utilization</div>
<p>Dynamic request scheduling that adds and removes sequences on the fly. No wasted compute on padding. Requests start generating immediately.</p>
</div>
<div class="feature-card">
<h3>Speculative Decoding</h3>
<div class="feat-stat">Draft proposes, target verifies</div>
<p>Draft model acceleration. A smaller model proposes candidate tokens, verified by the target model in parallel. Mathematically equivalent output.</p>
</div>
<div class="feature-card">
<h3>Tensor Parallelism</h3>
<div class="feat-stat">Scale to any model size</div>
<p>Split model layers across multiple GPUs. Serve models larger than single GPU memory with minimal communication overhead. Linear scaling.</p>
</div>
<div class="feature-card">
<h3>ISQ (In-Situ Quantization)</h3>
<div class="feat-stat">No pre-processing needed</div>
<p>Quantize models on load. Load BF16 weights and serve at INT4/INT8 with a single flag. No separate quantization step or tool required.</p>
</div>
<div class="feature-card">
<h3>Prompt Caching</h3>
<div class="feat-stat">Reuse computed prefixes</div>
<p>Automatically cache and reuse KV computations for shared prompt prefixes across requests. Dramatically reduces time-to-first-token for repeated contexts.</p>
</div>
</div>
</div>
</section>
<!-- ============================================================ -->
<!-- 6. Supported Model Architectures (table) -->
<!-- ============================================================ -->
<section id="compat">
<div class="container">
<div class="section-label">Compatibility</div>
<h2 class="section-title">Supported model architectures</h2>
<p class="section-desc">60+ architectures across text, code, vision, and embedding tasks. All major quantization formats supported.</p>
<div class="table-wrap">
<table>
<thead>
<tr>
<th>Family</th>
<th>Models</th>
<th>Quantization</th>
<th>Context</th>
</tr>
</thead>
<tbody>
<tr>
<td>Zen</td>
<td>zen4, zen4-max, zen4-coder, zen4-mini, zen4-ultra</td>
<td>GGUF, BF16, FP16</td>
<td>Up to 262K</td>
</tr>
<tr>
<td>Llama</td>
<td>Llama 3.1, 3.2, 3.3, 4</td>
<td>GGUF, GPTQ, AWQ, EXL2</td>
<td>Up to 128K</td>
</tr>
<tr>
<td>Qwen</td>
<td>Qwen3, QwQ</td>
<td>GGUF, GPTQ</td>
<td>Up to 131K</td>
</tr>
<tr>
<td>Mistral</td>
<td>Mistral, Mixtral, Mamba</td>
<td>GGUF, EXL2</td>
<td>Up to 32K</td>
</tr>
<tr>
<td>DeepSeek</td>
<td>V2, V3, R1, Coder</td>
<td>GGUF, GPTQ</td>
<td>Up to 128K</td>
</tr>
<tr>
<td>Gemma</td>
<td>Gemma 2, Gemma 3</td>
<td>GGUF</td>
<td>Up to 8K</td>
</tr>
<tr>
<td>Phi</td>
<td>Phi-3, Phi-4, Phi-MoE</td>
<td>GGUF</td>
<td>Up to 128K</td>
</tr>
<tr>
<td>Vision</td>
<td>Phi-3V, LLaVA, Idefics 3</td>
<td>GGUF</td>
<td>Up to 128K</td>
</tr>
<tr>
<td>Embedding</td>
<td>GTE, E5, BGE</td>
<td>FP16</td>
<td>Up to 8K</td>
</tr>
</tbody>
</table>
</div>
</div>
</section>
<!-- ============================================================ -->
<!-- 7. API Reference -->
<!-- ============================================================ -->
<section id="api">
<div class="container">
<div class="section-label">API</div>
<h2 class="section-title">OpenAI-compatible endpoints</h2>
<p class="section-desc">Drop-in replacement for the OpenAI API. Use existing client libraries, tools, and integrations with zero changes.</p>
<div class="endpoint-list">
<div class="endpoint">
<span class="method">POST</span>
<span class="path">/v1/chat/completions</span>
<span class="ep-desc">Chat with streaming support</span>
</div>
<div class="endpoint">
<span class="method">POST</span>
<span class="path">/v1/completions</span>
<span class="ep-desc">Text completions</span>
</div>
<div class="endpoint">
<span class="method">POST</span>
<span class="path">/v1/embeddings</span>
<span class="ep-desc">Vector embeddings</span>
</div>
<div class="endpoint">
<span class="method">GET</span>
<span class="path">/v1/models</span>
<span class="ep-desc">List loaded models</span>
</div>
<div class="endpoint">
<span class="method">GET</span>
<span class="path">/health</span>
<span class="ep-desc">Health check</span>
</div>
</div>
<div class="api-examples">
<div>
<div class="api-example-label">curl</div>
<div class="code-block standalone">
<pre><span class="cmd">curl http://localhost:8000/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "zenlm/zen4-mini",
"messages": [
{"role": "user", "content": "Hello"}
],
"stream": true
}'</span></pre>
</div>
</div>
<div>
<div class="api-example-label">Python</div>
<div class="code-block standalone">
<pre><span class="key">from</span> <span class="cmd">openai</span> <span class="key">import</span> <span class="cmd">OpenAI</span>
<span class="cmd">client</span> = <span class="cmd">OpenAI</span>(
<span class="key">base_url</span>=<span class="str">"http://localhost:8000/v1"</span>,
<span class="key">api_key</span>=<span class="str">"not-needed"</span>
)
<span class="cmd">response</span> = <span class="cmd">client.chat.completions.create</span>(
<span class="key">model</span>=<span class="str">"zenlm/zen4-mini"</span>,
<span class="key">messages</span>=[
{<span class="str">"role"</span>: <span class="str">"user"</span>,
<span class="str">"content"</span>: <span class="str">"Hello"</span>}
]
)
<span class="key">print</span>(<span class="cmd">response.choices</span>[0]<span class="cmd">.message.content</span>)</pre>
</div>
</div>
</div>
</div>
</section>
<!-- ============================================================ -->
<!-- 8. Comparison: Edge vs Engine -->
<!-- ============================================================ -->
<section id="comparison">
<div class="container">
<div class="section-label">Edge vs Engine</div>
<h2 class="section-title">Choose your runtime</h2>
<p class="section-desc">Hanzo Engine is for cloud servers and production API serving. Hanzo Edge is for on-device and in-browser inference. Same model format, different deployment targets.</p>
<div class="table-wrap comparison-table">
<table>
<thead>
<tr>
<th></th>
<th>Hanzo Edge</th>
<th>Hanzo Engine</th>
</tr>
</thead>
<tbody>
<tr>
<td>Target</td>
<td>Devices, browsers</td>
<td>Cloud servers</td>
</tr>
<tr>
<td>Models</td>
<td>GGUF quantized</td>
<td>All formats</td>
</tr>
<tr>
<td>GPU</td>
<td>Metal, CPU, WASM</td>
<td>CUDA, Metal, CPU</td>
</tr>
<tr>
<td>Batching</td>
<td>Single request</td>
<td>Continuous batching</td>
</tr>
<tr>
<td>Scale</td>
<td>Single user</td>
<td>Production serving</td>
</tr>
<tr>
<td>Install</td>
<td><code>cargo install hanzo-edge</code></td>
<td><code>cargo install hanzo-engine</code></td>
</tr>
</tbody>
</table>
</div>
</div>
</section>
<!-- ============================================================ -->
<!-- 10. Footer -->
<!-- ============================================================ -->
<footer>
<div class="container">
<span class="footer-copy">&copy; 2026 Hanzo AI Inc.</span>
<ul class="footer-links">
<li><a href="https://hanzoai.github.io/hanzo/benchmarks/prefill-by-prompt-length/">Prefill</a></li>
<li><a href="https://hanzoai.github.io/hanzo/benchmarks/decode-throughput/">Decode</a></li>
<li><a href="https://hanzoai.github.io/hanzo/benchmarks/time-to-first-token/">Time to first token</a></li>
<li><a href="https://hanzoai.github.io/hanzo/benchmarks/four-client-concurrency/">Concurrency</a></li>
<li><a href="https://hanzoai.github.io/hanzo/benchmarks/long-context-124k/">Long context</a></li>
<li><a href="https://github.com/hanzoai/engine">GitHub</a></li>
<li><a href="https://github.com/hanzoai/engine/pkgs/container/engine">Docker</a></li>
<li><a href="https://crates.io/crates/hanzo-engine">Crates.io</a></li>
<li><a href="https://hanzo.ai">Hanzo AI</a></li>
<li><a href="https://edge.hanzo.ai">Hanzo Edge</a></li>
</ul>
</div>
</footer>
<script>
function copyInstall(e) {
e.preventDefault();
var text = 'curl -sSL https://engine.hanzo.ai/install.sh | sh';
navigator.clipboard.writeText(text).then(function() {
var hint = document.querySelector('.copy-hint');
if (hint) {
hint.textContent = 'copied';
setTimeout(function() { hint.textContent = 'click to copy'; }, 2000);
}
});
}
function switchTab(e, panelId) {
var tabs = document.querySelectorAll('.deploy-tab');
var panels = document.querySelectorAll('.deploy-panel');
for (var i = 0; i < tabs.length; i++) {
tabs[i].classList.remove('active');
}
for (var i = 0; i < panels.length; i++) {
panels[i].classList.remove('active');
}
e.target.classList.add('active');
document.getElementById(panelId).classList.add('active');
}
</script>
</body>
</html>