Switch Ollama to gemma3n:e4b on node-197 GPU

Bake-off results: gemma3n:e4b (80.6% cmd match, 100% safety, 5.9s)
outperforms qwen3-coder:30b (67.7%, 93.5%, 14.7s) on all metrics.
Moved from steel141 CPU inference to node-197 RTX 4000 GPU.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

CONTEXT.md

@@ -113,3 +113,12 @@ For shrink-world use port `25576` and password `REDACTED_RCON`.
 - External access requires port forwarding on router: `25565` and `25566` → `192.168.0.244`
 - Web panel accessible via Caddy at `mc.sethpc.xyz`
 - DNS: Pi-hole at `192.168.0.153`
+
+---
+
+## AI / Ollama
+
+- **Ollama instance:** `192.168.0.179:11434` (CT 105, node-197, Quadro RTX 4000 8GB)
+- **Model (both message + command):** `gemma3n:e4b` (6.9B, Q4_K_M, GPU-accelerated)
+- **Previous:** `192.168.0.141:11434` (steel141), `gemma3:12b` + `qwen3-coder:30b`
+- **Changed:** 2026-03-18 after bake-off showed gemma3n:e4b outperforms qwen3-coder:30b (80.6% vs 67.7% cmd accuracy, 100% vs 93.5% safety, 3x faster)

mc_aigod_shrink.json

@@ -5,9 +5,9 @@
   "rcon_host": "127.0.0.1",
   "rcon_port": 25576,
   "rcon_password": "REDACTED_RCON",
-  "ollama_url": "http://192.168.0.141:11434",
-  "model": "gemma3:12b",
-  "command_model": "qwen3-coder:30b",
+  "ollama_url": "http://192.168.0.179:11434",
+  "model": "gemma3n:e4b",
+  "command_model": "gemma3n:e4b",
   "temperature": 0.85,
   "max_tokens": 600,
   "cooldown_seconds": 20,