feat(fleet): promote Ollama to first-class provider, assign Gemma 4 across fleet

- lazarus-registry.yaml: replace big_brain/RunPod with local ollama/gemma4:12b
- fleet-routing.json: assign ollama:gemma4:12b to carnice, bilbobagginshire, substratum
- intelligence/deepdive/config.yaml: local model -> gemma4:12b

fleet/fleet-routing.json

@@ -9,7 +9,7 @@
       "id": 27,
       "name": "carnice",
       "gitea_user": "carnice",
-      "model": "qwen3.5-9b",
+      "model": "ollama:gemma4:12b",
       "tier": "free",
       "location": "Local Metal",
       "description": "Local Hermes agent, fine-tuned on Hermes traces. Runs on local hardware.",
@@ -41,7 +41,7 @@
       "id": 25,
       "name": "bilbobagginshire",
       "gitea_user": "bilbobagginshire",
-      "model": "ollama",
+      "model": "ollama:gemma4:12b",
       "tier": "free",
       "location": "Bag End, The Shire (VPS)",
       "description": "Ollama on VPS. Speaks when spoken to. Prefers quiet. Not for delegated work.",
@@ -74,7 +74,7 @@
       "id": 23,
       "name": "substratum",
       "gitea_user": "substratum",
-      "model": "unassigned",
+      "model": "ollama:gemma4:12b",
       "tier": "unknown",
       "location": "Below the Surface",
       "description": "Infrastructure, deployments, bedrock services. Needs model assignment before activation.",

intelligence/deepdive/config.yaml

@@ -76,7 +76,7 @@ deepdive:
   # Phase 3: Synthesis
   synthesis:
     llm_endpoint: "http://localhost:4000/v1"  # Local llama-server
-    llm_model: "gemma-4-it"
+    llm_model: "gemma4:12b"
     max_summary_length: 800
     temperature: 0.7
 

lazarus-registry.yaml

@@ -20,8 +20,8 @@ fleet:
     - provider: openrouter
       model: anthropic/claude-sonnet-4-20250514
       timeout: 120
-    - provider: big_brain
-      model: gemma3:27b-instruct-q8_0
+    - provider: ollama
+      model: gemma4:12b
       timeout: 300
     health_endpoints:
       gateway: http://127.0.0.1:8646
@@ -100,11 +100,11 @@ provider_health_matrix:
     last_checked: '2026-04-07T02:55:00Z'
     rate_limited: false
     dead: false
-  big_brain:
-    status: provisioning
-    note: RunPod L40S instance big-brain-bezalel deployed; Ollama endpoint propagating
-    last_checked: '2026-04-07T02:55:00Z'
-    endpoint: http://yxw29g3excyddq-64411cd0-11434.tcp.runpod.net:11434/v1
+  ollama:
+    status: healthy
+    note: Local Ollama endpoint with Gemma 4 support
+    last_checked: '2026-04-07T15:09:53.385047+00:00'
+    endpoint: http://localhost:11434/v1
     rate_limited: false
     dead: false
 timeout_policies: