Upload production/.env.minimal

production/.env.minimal

@@ -6,20 +6,29 @@
 CLOUDFLARE_API_KEY=sk-your-cloudflare-api-key
 CLOUDFLARE_ACCOUNT_ID=your-account-id-here
 
-# OPTIONAL but recommended: NVIDIA NIM (primary — faster)
+# OPTIONAL but recommended: NVIDIA NIM (primary — fastest cloud)
 NVIDIA_API_KEY=nvapi-your-nvidia-api-key
 
+# OPTIONAL: Google Gemini (tertiary fallback — great for coding/reasoning)
+GEMINI_API_KEY=your-gemini-api-key-here
+
 # =====================================
 # That's it. Everything else auto-configures:
-#   • Primary:  NIM (if API key set)
-#   • Fallback: Cloudflare Workers AI
-#   • Cache:    Redis (auto-launched)
-#   • DB:       Postgres (auto-launched)
-#   • Metrics:  Prometheus + Grafana
-#   • MLX:      Off by default (enable below for local M2)
+#   • Primary:    NIM (if API key set)
+#   • Secondary:  Cloudflare Workers AI
+#   • Tertiary:   Google Gemini
+#   • Local:      MLX (off by default — enable below)
+#   • Cache:      Redis (auto-launched)
+#   • DB:         Postgres (auto-launched)
+#   • Metrics:    Prometheus + Grafana
 # =====================================
 
 # ─── Optional: Enable MLX local on M2 Pro Max ───
-# Uncomment to run 70B models locally via Metal GPU:
+# For Gemma 4 local (best Claude Opus alternative):
 # MLX_ENABLED=true
 # MLX_API_BASE=http://host.docker.internal:8000/v1
+#
+# Recommended Gemma 4 MLX models for 96GB:
+#   • gemma-4-31b-bf16  (~65GB runtime) — highest quality, best Opus alternative
+#   • gemma-4-26b-a4b-it-bf16 (~55GB) — slightly smaller, still excellent
+#   • gemma-4-e4b-it (~12GB) — fast, efficient, good for quick tasks