diff --git a/host_vars/artemis.yml b/host_vars/artemis.yml
index a20d6eb..51588b3 100644
--- a/host_vars/artemis.yml
+++ b/host_vars/artemis.yml
@@ -29,3 +29,6 @@ hermes_config:
   provider: openrouter
   model: openai/gpt-4o-mini
   context_length: 128000
+
+# Hermes auto-update offset (minutes from hour mark)
+hermes_update_offset: 0
diff --git a/host_vars/bones.yml b/host_vars/bones.yml
index d7eefbd..9a92de6 100644
--- a/host_vars/bones.yml
+++ b/host_vars/bones.yml
@@ -25,3 +25,6 @@ ollama_models:
 # Node-specific vars
 bones_storage: "256GB SSD"
 jvm_heap: "512m"
+
+# Hermes auto-update offset (minutes from hour mark)
+hermes_update_offset: 30
diff --git a/host_vars/hulkbuster.yml b/host_vars/hulkbuster.yml
index 226f198..5ad5754 100644
--- a/host_vars/hulkbuster.yml
+++ b/host_vars/hulkbuster.yml
@@ -6,9 +6,11 @@ gpu_type: nvidia
 gpu_model: "RTX 4070"
 vram_mb: 12282
 
-# GPU-specific tools (not drivers — Mark44 uses proprietary NVIDIA package)
+# GPU-specific packages
 extra_packages:
-  - nvtop                     # GPU monitoring
+  - nvidia-driver             # NVIDIA driver (if not already installed via proprietary)
+  - cuda-toolkit              # CUDA toolkit for ML workloads
+  - nvtop                     # GPU monitoring (better than nvidia-smi for live view)
   - nethogs                   # Network per-process monitoring
   - iotop                     # I/O per-process monitoring
 
@@ -17,6 +19,7 @@ ollama_models:
   - gemma4:e4b               # Already pulled — keep it
   - qwen2.5-coder:14b        # Primary coding model
   - qwen2.5:14b             # General purpose large model
+  - mistral:7b               # Fallback fast model
 
 # Services
 managed_services:
@@ -30,3 +33,6 @@ managed_services:
 # Ollama port override (standard)
 ollama_port: 11434
 open_webui_port: 8080
+
+# Hermes auto-update offset (minutes from hour mark)
+hermes_update_offset: 10
diff --git a/host_vars/mark5.yml b/host_vars/mark5.yml
index 1d4df70..a3d32e8 100644
--- a/host_vars/mark5.yml
+++ b/host_vars/mark5.yml
@@ -28,3 +28,6 @@ managed_services:
 
 # Ollama port
 ollama_port: 11434
+
+# Hermes auto-update offset (minutes from hour mark)
+hermes_update_offset: 20
diff --git a/host_vars/nebuchadnezzar.yml b/host_vars/nebuchadnezzar.yml
new file mode 100644
index 0000000..0c68b9b
--- /dev/null
+++ b/host_vars/nebuchadnezzar.yml
@@ -0,0 +1,3 @@
+
+# Hermes auto-update offset (minutes from hour mark)
+hermes_update_offset: 40
diff --git a/local.yml b/local.yml
index 4777995..1fe860d 100644
--- a/local.yml
+++ b/local.yml
@@ -89,6 +89,31 @@
         - ansible_os_family == "Debian"
       tags: [artemis]
 
+    # --- Hermes auto-update cron (fleet-wide, staggered offsets) ---
+    - name: Ensure hermes-update cron is installed
+      cron:
+        name: "hermes auto-update"
+        minute: "{{ hermes_update_offset | default(0) }}"
+        hour: "*/8"
+        user: root
+        job: "bash -c 'echo === $(date -Iseconds) hermes update === >> /var/log/hermes-update.log 2>&1 && hermes update --y >> /var/log/hermes-update.log 2>&1'"
+      when:
+        - ansible_os_family == "Debian"
+        - hermes_update_offset is defined
+      tags: [hermes_update]
+
+    - name: Ensure hermes-update log file exists
+      file:
+        path: /var/log/hermes-update.log
+        state: touch
+        owner: root
+        group: root
+        mode: '0666'
+      when:
+        - ansible_os_family == "Debian"
+        - hermes_update_offset is defined
+      tags: [hermes_update]
+
     - name: Print completion message
       debug:
         msg: "Baseline complete on {{ ansible_hostname }} — node_type={{ node_type | default('unspecified') }}, gpu={{ has_gpu | default(false) }}"