Bell card: link to the live demo at bell.entangledcode.dev

Author: dcoldeira

index.html

@@ -333,13 +333,16 @@ <h3><small>Forza7 · SailGenius</small>Marco, an LLM-powered passage planner</h3
       </div>
 
       <div class="card">
-        <h3><small>Independent · 2025</small>Bell, a self-hosted fine-tuned physics assistant</h3>
+        <h3><small>Independent · 2025–26</small>Bell, a quantum physics reasoning assistant</h3>
         <p>
-          Qwen3-8B with a LoRA adapter trained on 815 instruction pairs, as a provider-swappable
-          alternative to a hosted LLM API. Sandboxed Python execution, Docker/Caddy deployment,
-          GDPR-compliant. Ran live at bell.entangledcode.dev; now offline, code and training data kept.
+          Plain English in, real numbers out — questions about entanglement and causal structure
+          answered by generating and executing real QRL code, not by the model guessing a number.
+          Provider-swappable by design: running on Mistral (EU-hosted) today, a self-hosted
+          fine-tuned Qwen3-8B (815 training pairs) in standby. Sandboxed execution, Podman/Caddy
+          deployment on Hetzner.
         </p>
         <div class="card-links">
+          <a href="https://bell.entangledcode.dev">Try it — bell.entangledcode.dev</a>
           <a href="mailto:david@entangledcode.dev?subject=Bell%20code">Code on request</a>
         </div>
       </div>