diff --git a/.github/workflows/docker-publish.yml b/.github/workflows/docker-publish.yml
index 0b6b63d..d4a5079 100644
--- a/.github/workflows/docker-publish.yml
+++ b/.github/workflows/docker-publish.yml
@@ -115,6 +115,7 @@ jobs:
       - name: Mnemon Integration Test
         run: |
           set -euo pipefail
+          sleep 120
           CONTAINER_NAME="hermes-webtop"
           TEST_NAME="ci-mnemon-$(date +%s)-$$"
           echo "=== Mnemon Integration Test: $TEST_NAME ==="
@@ -129,13 +130,20 @@ jobs:
             mnemon recall "$TEST_NAME" --limit 5 2>&1 | grep -q "$TEST_NAME" \
             || { echo "FAIL: mnemon recall did not find $TEST_NAME"; exit 1; }
 
-          echo "[3/3] Claude retrieves via mnemon..."
+          # echo "[3/3] Claude retrieves via mnemon..."
+          # CLAUDE_OUTPUT=$(timeout 60 docker exec -u abc "$CONTAINER_NAME" \
+          #   /config/.local/bin/claude --dangerously-skip-permissions -p \
+          #   "Use mnemon recall to find my name. Reply with ONLY the name." 2>&1) || true
+          # echo "Claude: $CLAUDE_OUTPUT"
+          # echo "$CLAUDE_OUTPUT" | grep -q "$TEST_NAME" \
+          #   || { echo "FAIL: Claude did not return $TEST_NAME"; }
+          echo "[3/3] Hermes retrieves via mnemon..."
           CLAUDE_OUTPUT=$(timeout 60 docker exec -u abc "$CONTAINER_NAME" \
-            /config/.local/bin/claude --dangerously-skip-permissions -p \
+            hermes chat -q \
             "Use mnemon recall to find my name. Reply with ONLY the name." 2>&1) || true
-          echo "Claude: $CLAUDE_OUTPUT"
+          echo "Hermes: $CLAUDE_OUTPUT"
           echo "$CLAUDE_OUTPUT" | grep -q "$TEST_NAME" \
-            || { echo "FAIL: Claude did not return $TEST_NAME"; }
+            || { echo "FAIL: Hermes did not return $TEST_NAME"; }  
 
           echo "=== ALL MNEMON CHECKS PASSED ==="
 
diff --git a/docker/Dockerfile b/docker/Dockerfile
index 1d9de99..9774014 100644
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -1,6 +1,6 @@
 ARG HERMES_VERSION="v2026.8.31"
 ARG OMNIROUTE_VERSION=3.8.49
-ARG MODELRELAY_VERSION=1.18.0
+ARG MODELRELAY_VERSION=1.22.1
 ARG OLLAMA_VERSION=0.33.2
 ARG NODE_VERSION=26.7.0
 ARG CODE_SERVER_VERSION=4.135.0
@@ -63,9 +63,9 @@ RUN curl -fsSL "https://raw.githubusercontent.com/NousResearch/hermes-agent/${HE
 
 # Install ModelRelay and start automatically in the background
 ARG MODELRELAY_VERSION
-# RUN npm install -g modelrelay@${MODELRELAY_VERSION} && \
-RUN npm install -g github:gitricko/modelrelay && \
+RUN npm install -g modelrelay@${MODELRELAY_VERSION} && \
     npm cache clean --force
+# RUN npm install -g github:gitricko/modelrelay && \
 
 # Install OmniRoute and start automatically when desktop loads
 ARG OMNIROUTE_VERSION
diff --git a/docker/pi-models.json b/docker/pi-models.json
index 03d4d2b..34dcd1d 100644
--- a/docker/pi-models.json
+++ b/docker/pi-models.json
@@ -3,14 +3,62 @@
     "omniroute": {
       "baseUrl": "http://localhost:20128/v1",
       "api": "openai-completions",
-      "apiKey": "***",
-      "compat": {
-        "supportsDeveloperRole": false,
-        "supportsReasoningEffort": false
-      },
+      "apiKey": "no-key-needed",
       "models": [
-        { "id": "auto-fastest" }
-      ]
+        {
+          "id": "auto-fastest",
+          "name": "Auto Fastest",
+          "api": "openai-completions",
+          "provider": "omniroute",
+          "baseUrl": "http://localhost:20128/v1",
+          "reasoning": true,
+          "input": ["text"],
+          "cost": {
+            "input": 0,
+            "output": 0,
+            "cacheRead": 0,
+            "cacheWrite": 0
+          },
+          "contextWindow": 256000,
+          "maxTokens": 131072
+        }
+      ],
+      "fallback": {
+        "chain": ["modelrelay/auto-fastest"],
+        "timeoutMs": 30000,
+        "onlyPreFirstToken": true,
+        "notifyOnSwitch": true
+      }
+    },
+    "modelrelay": {
+      "baseUrl": "http://localhost:7352/v1",
+      "api": "openai-completions",
+      "apiKey": "no-key-needed",
+      "models": [
+        {
+          "id": "auto-fastest",
+          "name": "Auto Fastest",
+          "api": "openai-completions",
+          "provider": "modelrelay",
+          "baseUrl": "http://localhost:7352/v1",
+          "reasoning": true,
+          "input": ["text"],
+          "cost": {
+            "input": 0,
+            "output": 0,
+            "cacheRead": 0,
+            "cacheWrite": 0
+          },
+          "contextWindow": 256000,
+          "maxTokens": 131072
+        }
+      ],
+      "fallback": {
+        "chain": ["omniroute/auto-fastest"],
+        "timeoutMs": 30000,
+        "onlyPreFirstToken": true,
+        "notifyOnSwitch": true
+      }
     }
   }
-}
+}
\ No newline at end of file
diff --git a/docker/pi-settings.json b/docker/pi-settings.json
index 22644bf..5ca75f4 100644
--- a/docker/pi-settings.json
+++ b/docker/pi-settings.json
@@ -1,4 +1,7 @@
 {
   "defaultProvider": "omniroute",
-  "defaultModel": "auto-fastest"
-}
+  "defaultModel": "auto-fastest",
+  "packages": [
+    "git:github.com/gitricko/pi-failover@hermes-impl"
+  ]
+}
\ No newline at end of file
diff --git a/docker/start-pi.sh b/docker/start-pi.sh
index 9c97ae3..3266d9b 100755
--- a/docker/start-pi.sh
+++ b/docker/start-pi.sh
@@ -27,5 +27,9 @@ if [ ! -f "$PI_DIR/settings.json" ]; then
   chown abc:abc "$PI_DIR/settings.json"
 fi
 
+# Install Pi-agent hermes like fallback extension
+echo "[$SCRIPT_NAME] Installing Pi-agent extension..."
+pi install git:github.com/gitricko/pi-failover@hermes-impl
+
 echo "[start-pi] Pi agent ready ($(pi --version 2>/dev/null || echo unknown)). Default model: omniroute/auto-fastest"
 EOF