diff --git a/services/hermes/agent-deployment.yaml b/services/hermes/agent-deployment.yaml
index 9fce2b2c..822cf8b9 100644
--- a/services/hermes/agent-deployment.yaml
+++ b/services/hermes/agent-deployment.yaml
@@ -24,7 +24,7 @@ spec:
         ai.bstein.dev/router-wire-contract: ollama-numeric-keepalive
         ai.bstein.dev/execution: Hermes Kanban with durable direct Codex and Claude Code CLI workers
         ai.bstein.dev/model-policy: Jetson-assisted AUTO routing, low through xhigh, cross-provider fallback
-        ai.bstein.dev/placement: titan-08 rpi5; storage-backbone nodes excluded
+        ai.bstein.dev/placement: primary amd64 accelerator titan-22; arm64 rpi5 fleet fallback; storage-backbone nodes excluded
         ai.bstein.dev/config-rev: "20260824-claude-fable-quota"
         prometheus.io/scrape: "true"
         prometheus.io/path: /metrics
@@ -97,10 +97,19 @@ spec:
       securityContext:
         seccompProfile:
           type: RuntimeDefault
+      # APPLY ONLY AFTER the multi-arch hermes-agent image is built + validated
+      # (both arch leaves + promoted index). titan-22 is the STRONG/primary home
+      # for hermes-agent; the arm64 pi fleet is an OR'd fallback so the worker is
+      # never stranded. titan-22 is a hard-tainted accelerator, so this pod does
+      # NOT require node-role.kubernetes.io/worker for titan-22 and instead
+      # tolerates the accelerator taint + targets the accelerator label.
       affinity:
         nodeAffinity:
           requiredDuringSchedulingIgnoredDuringExecution:
             nodeSelectorTerms:
+              # Fallback set: the arm64 pi fleet (unchanged). nodeSelectorTerms
+              # are OR'd, so hermes-agent still runs here if titan-22 is
+              # unavailable.
               - matchExpressions:
                   - key: kubernetes.io/arch
                     operator: In
@@ -111,13 +120,42 @@ spec:
                   - key: kubernetes.io/hostname
                     operator: NotIn
                     values: [titan-04, titan-06, titan-08, titan-13, titan-14, titan-17, titan-18, titan-19]
+              # Primary amd64 home: accelerator titan-22. No worker=true here.
+              - matchExpressions:
+                  - key: kubernetes.io/arch
+                    operator: In
+                    values: [amd64]
+                  - key: node-role.kubernetes.io/accelerator
+                    operator: In
+                    values: ["true"]
+                  - key: kubernetes.io/hostname
+                    operator: In
+                    values: [titan-22]
           preferredDuringSchedulingIgnoredDuringExecution:
+            # STRONG primary preference: titan-22 (weight 100, the scheduler
+            # maximum) outranks the pi-fleet nudge below, so hermes-agent
+            # actually lives on titan-22 whenever it is schedulable.
             - weight: 100
+              preference:
+                matchExpressions:
+                  - key: kubernetes.io/hostname
+                    operator: In
+                    values: [titan-22]
+            # Weaker fallback nudge within the pi fleet.
+            - weight: 50
               preference:
                 matchExpressions:
                   - key: hardware
                     operator: In
                     values: [rpi5]
+      tolerations:
+        # Permit (do not force) scheduling onto the hard-tainted accelerator
+        # titan-22. This only lets hermes-agent consider titan-22; it does not
+        # change jellyfin's media-core priority or preemption.
+        - key: node-role.kubernetes.io/accelerator
+          operator: Equal
+          value: "true"
+          effect: NoSchedule
       initContainers:
         - name: init-config
           image: busybox:1.37
diff --git a/testing/tests/test_hermes_agent_layout.py b/testing/tests/test_hermes_agent_layout.py
index 3c343e67..1d9de7cb 100644
--- a/testing/tests/test_hermes_agent_layout.py
+++ b/testing/tests/test_hermes_agent_layout.py
@@ -154,12 +154,37 @@ def test_agent_avoids_unhealthy_nodes_and_stays_on_storage_workers():
     assert hostnames["operator"] == "NotIn"
     assert set(hostnames["values"]) >= {"titan-04", "titan-19"}
 
-    assert len(terms) == 1
+    # Two OR'd terms: the arm64 pi fleet (fallback) and the amd64 accelerator
+    # titan-22 (primary home once the multi-arch image exists).
+    assert len(terms) == 2
+    pi_fleet = {item["key"]: item for item in terms[0]["matchExpressions"]}
+    assert pi_fleet["kubernetes.io/arch"]["values"] == ["arm64"]
+    assert pi_fleet["node-role.kubernetes.io/worker"]["values"] == ["true"]
+
+    titan22 = {item["key"]: item for item in terms[1]["matchExpressions"]}
+    assert titan22["kubernetes.io/arch"]["values"] == ["amd64"]
+    assert titan22["kubernetes.io/hostname"]["values"] == ["titan-22"]
+    assert titan22["node-role.kubernetes.io/accelerator"]["values"] == ["true"]
+    # titan-22 access must NOT depend on the generic worker label.
+    assert "node-role.kubernetes.io/worker" not in titan22
 
     preferences = pod["affinity"]["nodeAffinity"][
         "preferredDuringSchedulingIgnoredDuringExecution"
     ]
-    assert [item["weight"] for item in preferences] == [100]
+    # titan-22 is the STRONG primary (max weight 100) and outranks the pi-fleet
+    # nudge (50), so hermes-agent actually lives on titan-22 when schedulable.
+    assert [item["weight"] for item in preferences] == [100, 50]
+    primary = preferences[0]["preference"]["matchExpressions"][0]
+    assert primary["key"] == "kubernetes.io/hostname"
+    assert primary["values"] == ["titan-22"]
+
+    # The pod tolerates the accelerator hard taint so it can consider titan-22.
+    assert {
+        "key": "node-role.kubernetes.io/accelerator",
+        "operator": "Equal",
+        "value": "true",
+        "effect": "NoSchedule",
+    } in pod["tolerations"]
 
     hermes = next(item for item in pod["containers"] if item["name"] == "hermes")
     assert hermes["resources"]["requests"] == {