diff --git a/src/pages/index.astro b/src/pages/index.astro
index 26a027c..8b3b3b5 100644
--- a/src/pages/index.astro
+++ b/src/pages/index.astro
@@ -25,6 +25,8 @@ const version = pkg.version ?? "";
           <div class="hero-meta">
             <span class="meta-item">RUNS LOCAL</span>
             <span class="meta-divider">/</span>
+            <span class="meta-item">LOCAL AI</span>
+            <span class="meta-divider">/</span>
             <span class="meta-item">WHOLE PROJECT CONTEXT</span>
             <span class="meta-divider">/</span>
             <span class="meta-item">NO CLOUD REQUIRED</span>
@@ -32,6 +34,9 @@ const version = pkg.version ?? "";
           <p class="hero-description">
             Zaguán Blade + zcoderd are built for long-running coding sessions. In our 250-turn benchmark, context compression and selective retrieval reduced token cost by ~31% on average, with higher gains in context-heavy runs. It’s not always cheaper in lean sessions — it’s designed to prevent context bloat when history grows.
           </p>
+          <p class="hero-description" style="margin-top: var(--space-sm);">
+            Connect to local models via Ollama or any OpenAI-compatible endpoint. Your code, your AI, your hardware — fully private and offline capable.
+          </p>
           <div class="hero-actions">
             <a href="#download" class="action-primary">
               <span class="action-text">RUN AI LOCALLY. START HERE.</span>