Buy advice: mini-PC category now starts at a $249 Jetson, real cheap-to-big ladder

engine/real_advisor.py

@@ -759,9 +759,17 @@ _PLATFORMS = [
         ],
     },
     {
-        "key": "mini", "name": "Mini-PC (unified memory)", "icon": "monitor",
-        "blurb": "Big models in a small, silent, low-power box. They fit huge models; they run them at a steady jog, not a sprint.",
+        "key": "mini", "name": "Mini-PC / edge box", "icon": "monitor",
+        "blurb": "Small, silent, low-power boxes, from a $249 CUDA Jetson up to unified-memory mini-PCs that fit huge models at a steady jog, not a sprint.",
         "tiers": [
+            ("Jetson Orin Nano Super, 8 GB", "$249", dict(ram_gb=8, unified=True, vendor="nvidia"),
+             "NVIDIA Jetson Orin Nano Super dev kit, 8 GB unified LPDDR5, CUDA, 67 TOPS",
+             "A complete CUDA dev kit for $249, a price that has held through the 2026 shortage. 8 GB of unified memory caps you to small quantised models, but unlike a Raspberry Pi (which is CPU-only and, oddly, now costs more) it has a real GPU, so those models actually accelerate. The best dollar-per-CUDA box here, ideal for tinkering and tiny always-on models.",
+             "https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-orin/nano-super-developer-kit/"),
+            ("Jetson Orin NX, 16 GB", "$949", dict(ram_gb=16, unified=True, vendor="nvidia"),
+             "NVIDIA Jetson Orin NX 16 GB unified LPDDR5, CUDA, 100 TOPS (Seeed reComputer J4012)",
+             "The cheapest Jetson that crosses 16 GB of unified memory with CUDA, sold ready-to-run as the Seeed reComputer J4012 (module, carrier board and a 128 GB SSD) for about $949. Doubles the Orin Nano's memory, so it runs small-to-mid models with GPU acceleration, still well under a Ryzen AI Max+ box. Bare modules cost less but you assemble the rest.",
+             "https://www.seeedstudio.com/reComputer-J4012-p-5586.html"),
             ("Ryzen AI Max+ 395, 64 GB", "$1,499-1,700", dict(ram_gb=64, unified=True, vendor="amd"),
              "AMD Ryzen AI Max+ 395 'Strix Halo', 64 GB unified memory, ~256 GB/s",
              "16 Zen 5 cores plus a strong integrated GPU sharing 64 GB, in a tiny quiet box, no discrete card. Runs ~30B comfortably. Examples: GMKtec EVO-X2 around $1,499, Framework Desktop $1,639. Bandwidth is far below a discrete GPU, so big models fit and are usable but not fast. Soldered RAM, so buy the capacity up front.",