diff --git a/examples/backends/sglang/deploy/README.md b/examples/backends/sglang/deploy/README.md
index 44a94c44484d..6040c4865b32 100644
--- a/examples/backends/sglang/deploy/README.md
+++ b/examples/backends/sglang/deploy/README.md
@@ -9,6 +9,7 @@ Basic deployment pattern with frontend and a single decode worker.
 
 **Architecture:**
 - `Frontend`: OpenAI-compatible API server
+- `ModelExpress`: Shared model caching service across workers
 - `SGLangDecodeWorker`: Single worker handling both prefill and decode
 
 ### 2. **Aggregated Router Deployment** (`agg_router.yaml`)
@@ -16,13 +17,15 @@ Enhanced aggregated deployment with KV cache routing capabilities.
 
 **Architecture:**
 - `Frontend`: OpenAI-compatible API server with router mode enabled (`--router-mode kv`)
+- `ModelExpress`: Shared model caching service across workers
 - `SGLangDecodeWorker`: Single worker handling both prefill and decode
 
-### 3. **Disaggregated Deployment** (`disagg.yaml`)**
+### 3. **Disaggregated Deployment** (`disagg.yaml`)
 High-performance deployment with separated prefill and decode workers.
 
 **Architecture:**
 - `Frontend`: HTTP API server coordinating between workers
+- `ModelExpress`: Shared model caching service across workers
 - `SGLangDecodeWorker`: Specialized decode-only worker (`--disaggregation-mode decode`)
 - `SGLangPrefillWorker`: Specialized prefill-only worker (`--disaggregation-mode prefill`)
 - Communication via NIXL transfer backend (`--disaggregation-transfer-backend nixl`)
@@ -79,6 +82,28 @@ Before using these templates, ensure you have:
 3. **Container registry access** for SGLang runtime images
 4. **HuggingFace token secret** (referenced as `envFromSecret: hf-token-secret`)
 
+### Persistent Volume Claim (PVC)
+
+All templates expect a pre-created PVC named `model-cache-pvc` for the shared model cache used by ModelExpress and SGLang workers.
+
+Apply the shared PVC once per namespace before deploying any graph:
+
+```bash
+kubectl apply -f model_cache_pvc.yaml -n $NAMESPACE
+```
+
+Note: If your cluster requires a specific storage class, edit `model_cache_pvc.yaml` to set `storageClassName` accordingly.
+
+### Container Images
+
+We have public images available on [NGC Catalog](https://catalog.ngc.nvidia.com/orgs/nvidia/teams/ai-dynamo/collections/ai-dynamo/artifacts). If you'd prefer to use your own registry, build and push your own image:
+
+```bash
+./container/build.sh --framework sglang
+# Tag and push to your container registry
+# Update the image references in the YAML files
+```
+
 ## Usage
 
 ### 1. Choose Your Template
@@ -118,6 +143,11 @@ Then, deploy the model using the deployment file.
 
 ```bash
 export DEPLOYMENT_FILE=agg.yaml
+
+# Create the shared model cache PVC (run once per namespace)
+kubectl apply -f model_cache_pvc.yaml -n $NAMESPACE
+
+# Apply the SGLang deployment
 kubectl apply -f $DEPLOYMENT_FILE -n ${NAMESPACE}
 ```
 
@@ -135,7 +165,7 @@ kubectl apply -f $DEPLOYMENT_FILE.generated -n $NAMESPACE
 
 ## Model Configuration
 
-All templates use **DeepSeek-R1-Distill-Llama-8B** as the default model. But you can use any sglang argument and configuration. Key parameters:
+All templates use **Qwen/Qwen3-0.6B** as the default model. You can use any SGLang arguments and configuration. Key parameters include `--model-path`, `--served-model-name`, and disaggregation flags (see YAMLs for examples).
 
 ## Monitoring and Health
 
diff --git a/examples/backends/sglang/deploy/agg.yaml b/examples/backends/sglang/deploy/agg.yaml
index c6302906b6b6..eb4fb281e51b 100644
--- a/examples/backends/sglang/deploy/agg.yaml
+++ b/examples/backends/sglang/deploy/agg.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: sglang-agg
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://sglang-agg-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: sglang-agg
@@ -14,6 +24,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: my-registry/sglang-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: sglang-agg
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     decode:
       envFromSecret: hf-token-secret
       dynamoNamespace: sglang-agg
@@ -41,4 +103,7 @@ spec:
             - "1"
             - --trust-remote-code
             - --skip-tokenizer-init
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
 
diff --git a/examples/backends/sglang/deploy/agg_logging.yaml b/examples/backends/sglang/deploy/agg_logging.yaml
index a6e576aae932..93f6f90ea29b 100644
--- a/examples/backends/sglang/deploy/agg_logging.yaml
+++ b/examples/backends/sglang/deploy/agg_logging.yaml
@@ -9,6 +9,15 @@ spec:
   envs:
   - name: DYN_LOGGING_JSONL
     value: "1"
+  - name: MODEL_EXPRESS_CACHE_DIRECTORY
+    value: "/model/.model-express/cache"
+  - name: HF_HUB_CACHE
+    value: "/model/.model-express/cache"
+  - name: MODEL_EXPRESS_URL
+    value: "http://sglang-agg-modelexpress:8000"
+  pvcs:
+    - name: model-cache-pvc
+      create: false
   services:
     Frontend:
       dynamoNamespace: sglang-agg
@@ -17,6 +26,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: my-registry/sglang-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: sglang-agg
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     decode:
       envFromSecret: hf-token-secret
       dynamoNamespace: sglang-agg
@@ -43,4 +104,7 @@ spec:
             - --tp
             - "1"
             - --trust-remote-code
-            - --skip-tokenizer-init
\ No newline at end of file
+            - --skip-tokenizer-init
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
\ No newline at end of file
diff --git a/examples/backends/sglang/deploy/agg_router.yaml b/examples/backends/sglang/deploy/agg_router.yaml
index e9e01e8bb81d..e71965b50e2b 100644
--- a/examples/backends/sglang/deploy/agg_router.yaml
+++ b/examples/backends/sglang/deploy/agg_router.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: sglang-agg-router
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://sglang-agg-router-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: sglang-agg-router
@@ -17,6 +27,58 @@ spec:
       envs:
         - name: DYN_ROUTER_MODE
           value: kv
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: sglang-agg-router
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     decode:
       envFromSecret: hf-token-secret
       dynamoNamespace: sglang-agg-router
@@ -44,3 +106,6 @@ spec:
             - "1"
             - --trust-remote-code
             - --skip-tokenizer-init
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/sglang/deploy/disagg-multinode.yaml b/examples/backends/sglang/deploy/disagg-multinode.yaml
index ac161b3159e5..d6e2b749c852 100644
--- a/examples/backends/sglang/deploy/disagg-multinode.yaml
+++ b/examples/backends/sglang/deploy/disagg-multinode.yaml
@@ -14,7 +14,16 @@ spec:
         key: HF_TOKEN
   - name: GLOO_SOCKET_IFNAME
     value: "eth0"
+  - name: MODEL_EXPRESS_CACHE_DIRECTORY
+    value: "/model/.model-express/cache"
+  - name: HF_HUB_CACHE
+    value: "/model/.model-express/cache"
+  - name: MODEL_EXPRESS_URL
+    value: "http://sglang-disagg-multinode-modelexpress:8000"
   backendFramework: sglang
+  pvcs:
+    - name: model-cache-pvc
+      create: false
   services:
     Frontend:
       dynamoNamespace: sglang-disagg-multinode
@@ -23,6 +32,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: my-registry/sglang-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: sglang-disagg-multinode
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     decode:
       multinode:
         nodeCount: 2
@@ -60,6 +121,9 @@ spec:
             - "0.0.0.0"
             - --mem-fraction-static
             - "0.82"
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     prefill:
       multinode:
         nodeCount: 2
@@ -96,4 +160,7 @@ spec:
             - --mem-fraction-static
             - "0.82"
             - --host
-            - "0.0.0.0"
\ No newline at end of file
+            - "0.0.0.0"
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
\ No newline at end of file
diff --git a/examples/backends/sglang/deploy/disagg.yaml b/examples/backends/sglang/deploy/disagg.yaml
index 52866fedf56a..93ac6ad21728 100644
--- a/examples/backends/sglang/deploy/disagg.yaml
+++ b/examples/backends/sglang/deploy/disagg.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: sglang-disagg
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://sglang-disagg-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: sglang-disagg
@@ -14,6 +24,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: my-registry/sglang-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: sglang-disagg
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     decode:
       envFromSecret: hf-token-secret
       dynamoNamespace: sglang-disagg
@@ -50,6 +112,9 @@ spec:
             - "12345"
             - --host
             - "0.0.0.0"
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     prefill:
       envFromSecret: hf-token-secret
       dynamoNamespace: sglang-disagg
@@ -85,4 +150,7 @@ spec:
             - --disaggregation-bootstrap-port
             - "12345"
             - --host
-            - "0.0.0.0"
\ No newline at end of file
+            - "0.0.0.0"
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
\ No newline at end of file
diff --git a/examples/backends/sglang/deploy/disagg_planner.yaml b/examples/backends/sglang/deploy/disagg_planner.yaml
index 1ed1d195ba49..f9fbe735d280 100644
--- a/examples/backends/sglang/deploy/disagg_planner.yaml
+++ b/examples/backends/sglang/deploy/disagg_planner.yaml
@@ -9,6 +9,15 @@ spec:
   pvcs:
     - name: dynamo-pvc
       create: false # Must be pre-created before deployment and SLA profiler must have been run
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://sglang-disagg-planner-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: dynamo
@@ -17,6 +26,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: my-registry/sglang-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: dynamo
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     Planner:
       dynamoNamespace: dynamo
       envFromSecret: hf-token-secret
@@ -74,6 +135,9 @@ spec:
             - "12345"
             - --host
             - "0.0.0.0"
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     prefill:
       dynamoNamespace: dynamo
       envFromSecret: hf-token-secret
@@ -110,3 +174,6 @@ spec:
             - "12345"
             - --host
             - "0.0.0.0"
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/sglang/deploy/model_cache_pvc.yaml b/examples/backends/sglang/deploy/model_cache_pvc.yaml
new file mode 100644
index 000000000000..56ad5f005224
--- /dev/null
+++ b/examples/backends/sglang/deploy/model_cache_pvc.yaml
@@ -0,0 +1,16 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: model-cache-pvc
+spec:
+  accessModes:
+    - ReadWriteMany
+  storageClassName: csi-mounted-fs-path-sc
+  resources:
+    requests:
+      storage: 256Gi
+    limits:
+      storage: 256Gi
diff --git a/examples/backends/trtllm/deploy/README.md b/examples/backends/trtllm/deploy/README.md
index 0f7aecd9f2f0..7ba16d40ae7d 100644
--- a/examples/backends/trtllm/deploy/README.md
+++ b/examples/backends/trtllm/deploy/README.md
@@ -9,6 +9,7 @@ Basic deployment pattern with frontend and a single worker.
 
 **Architecture:**
 - `Frontend`: OpenAI-compatible API server (with kv router mode disabled)
+- `ModelExpress`: Shared model caching service across workers
 - `TRTLLMWorker`: Single worker handling both prefill and decode
 
 ### 2. **Aggregated Router Deployment** (`agg_router.yaml`)
@@ -16,6 +17,7 @@ Enhanced aggregated deployment with KV cache routing capabilities.
 
 **Architecture:**
 - `Frontend`: OpenAI-compatible API server (with kv router mode enabled)
+- `ModelExpress`: Shared model caching service across workers
 - `TRTLLMWorker`: Multiple workers handling both prefill and decode (2 replicas for load balancing)
 
 ### 3. **Disaggregated Deployment** (`disagg.yaml`)
@@ -23,6 +25,7 @@ High-performance deployment with separated prefill and decode workers.
 
 **Architecture:**
 - `Frontend`: HTTP API server coordinating between workers
+- `ModelExpress`: Shared model caching service across workers
 - `TRTLLMDecodeWorker`: Specialized decode-only worker
 - `TRTLLMPrefillWorker`: Specialized prefill-only worker
 
@@ -31,6 +34,7 @@ Advanced disaggregated deployment with KV cache routing capabilities.
 
 **Architecture:**
 - `Frontend`: HTTP API server (with kv router mode enabled)
+- `ModelExpress`: Shared model caching service across workers
 - `TRTLLMDecodeWorker`: Specialized decode-only worker
 - `TRTLLMPrefillWorker`: Specialized prefill-only worker (2 replicas for load balancing)
 
@@ -40,6 +44,7 @@ Aggregated deployment with custom configuration.
 **Architecture:**
 - `nvidia-config`: ConfigMap containing a custom trtllm configuration
 - `Frontend`: OpenAI-compatible API server (with kv router mode disabled)
+- `ModelExpress`: Shared model caching service across workers
 - `TRTLLMWorker`: Single worker handling both prefill and decode with custom configuration mounted from the configmap
 
 ### 6. **Disaggregated Planner Deployment** (`disagg_planner.yaml`)
@@ -47,6 +52,7 @@ Advanced disaggregated deployment with SLA-based automatic scaling.
 
 **Architecture:**
 - `Frontend`: HTTP API server coordinating between workers
+- `ModelExpress`: Shared model caching service across workers
 - `Planner`: SLA-based planner that monitors performance and scales workers automatically
 - `Prometheus`: Metrics collection and monitoring
 - `TRTLLMDecodeWorker`: Specialized decode-only worker
@@ -107,6 +113,16 @@ Before using these templates, ensure you have:
 3. **Container registry access** for TensorRT-LLM runtime images
 4. **HuggingFace token secret** (referenced as `envFromSecret: hf-token-secret`)
 
+### Persistent Volume Claim (PVC)
+
+All templates expect a pre-created PVC named `model-cache-pvc` for the shared model cache used by ModelExpress and TensorRT-LLM workers.
+
+Apply the shared PVC once per namespace before deploying any graph:
+
+```bash
+kubectl apply -f model_cache_pvc.yaml -n $NAMESPACE
+```
+
 ### Container Images
 
 The deployment files currently require access to `my-registry/trtllm-runtime`. If you don't have access, build and push your own image:
@@ -170,6 +186,11 @@ Export the NAMESPACE you used in your Dynamo Cloud Installation.
 ```bash
 cd dynamo/examples/backends/trtllm/deploy
 export DEPLOYMENT_FILE=agg.yaml
+
+# Create the shared model cache PVC (run once per namespace)
+kubectl apply -f model_cache_pvc.yaml -n $NAMESPACE
+
+# Apply the deployment
 kubectl apply -f $DEPLOYMENT_FILE -n $NAMESPACE
 ```
 
diff --git a/examples/backends/trtllm/deploy/agg-with-config.yaml b/examples/backends/trtllm/deploy/agg-with-config.yaml
index d18d1b0fb2fb..a263c3bb5de4 100644
--- a/examples/backends/trtllm/deploy/agg-with-config.yaml
+++ b/examples/backends/trtllm/deploy/agg-with-config.yaml
@@ -27,6 +27,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: trtllm-agg
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://trtllm-agg-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: trtllm-agg
@@ -35,6 +45,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: nvcr.io/nvidia/ai-dynamo/tensorrtllm-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: trtllm-agg
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     TRTLLMWorker:
       envFromSecret: hf-token-secret
       dynamoNamespace: trtllm-agg
@@ -68,3 +130,6 @@ spec:
             - Qwen/Qwen3-0.6B
             - --extra-engine-args
             - ./recipes/qwen3/trtllm/agg.yaml
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/trtllm/deploy/agg.yaml b/examples/backends/trtllm/deploy/agg.yaml
index 54412576a258..616fd3dae7c3 100644
--- a/examples/backends/trtllm/deploy/agg.yaml
+++ b/examples/backends/trtllm/deploy/agg.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: trtllm-agg
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://trtllm-agg-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: trtllm-agg
@@ -14,6 +24,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: my-registry/trtllm-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: trtllm-agg
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     TRTLLMWorker:
       envFromSecret: hf-token-secret
       dynamoNamespace: trtllm-agg
@@ -37,3 +99,6 @@ spec:
             - Qwen/Qwen3-0.6B
             - --extra-engine-args
             - ./recipes/qwen3/trtllm/agg.yaml
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/trtllm/deploy/agg_router.yaml b/examples/backends/trtllm/deploy/agg_router.yaml
index ed42129fb421..7e438764dfcb 100644
--- a/examples/backends/trtllm/deploy/agg_router.yaml
+++ b/examples/backends/trtllm/deploy/agg_router.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: trtllm-agg-router
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://trtllm-agg-router-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: trtllm-agg-router
@@ -17,6 +27,58 @@ spec:
       envs:
         - name: DYN_ROUTER_MODE
           value: kv
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: trtllm-agg-router
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     TRTLLMWorker:
       envFromSecret: hf-token-secret
       dynamoNamespace: trtllm-agg-router
@@ -41,3 +103,6 @@ spec:
             - --extra-engine-args
             - ./recipes/qwen3/trtllm/agg.yaml
             - --publish-events-and-metrics
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/trtllm/deploy/disagg-multinode.yaml b/examples/backends/trtllm/deploy/disagg-multinode.yaml
index 6ba847157379..5dec109e66c3 100644
--- a/examples/backends/trtllm/deploy/disagg-multinode.yaml
+++ b/examples/backends/trtllm/deploy/disagg-multinode.yaml
@@ -62,17 +62,6 @@ data:
     cache_transceiver_config:
       backend: DEFAULT
 ---
-apiVersion: v1
-kind: PersistentVolumeClaim
-metadata:
-  name: models
-spec:
-  accessModes:
-    - ReadWriteMany
-  resources:
-    requests:
-      storage: 100Gi
----
 apiVersion: nvidia.com/v1alpha1
 kind: DynamoGraphDeployment
 metadata:
@@ -80,14 +69,19 @@ metadata:
 spec:
   backendFramework: trtllm
   pvcs:
-    - name: models
+    - name: model-cache-pvc
+      create: false
   envs:
     - name: OMPI_ALLOW_RUN_AS_ROOT
       value: "1"
     - name: OMPI_ALLOW_RUN_AS_ROOT_CONFIRM
       value: "1"
-    - name: HF_HOME
-      value: "/models"
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://trtllm-disagg-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: trtllm-disagg
@@ -104,10 +98,62 @@ spec:
           args:
             - --http-port
             - "8000"
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: trtllm-disagg
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     prefill:
       volumeMounts:
-        - name: models
-          mountPoint: /models
+        - name: model-cache-pvc
+          mountPoint: /model
       dynamoNamespace: trtllm-disagg
       envFromSecret: hf-token-secret
       componentType: worker
@@ -146,8 +192,8 @@ spec:
             - decode_first
     decode:
       volumeMounts:
-        - name: models
-          mountPoint: /models
+        - name: model-cache-pvc
+          mountPoint: /model
       dynamoNamespace: trtllm-disagg
       envFromSecret: hf-token-secret
       componentType: worker
diff --git a/examples/backends/trtllm/deploy/disagg.yaml b/examples/backends/trtllm/deploy/disagg.yaml
index 501d2a4c200e..578708c36b77 100644
--- a/examples/backends/trtllm/deploy/disagg.yaml
+++ b/examples/backends/trtllm/deploy/disagg.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: trtllm-disagg
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://trtllm-disagg-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: trtllm-disagg
@@ -14,6 +24,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: my-registry/trtllm-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: trtllm-disagg
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     TRTLLMPrefillWorker:
       dynamoNamespace: trtllm-disagg
       envFromSecret: hf-token-secret
@@ -42,6 +104,9 @@ spec:
             - prefill
             - --disaggregation-strategy
             - decode_first
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     TRTLLMDecodeWorker:
       dynamoNamespace: trtllm-disagg
       envFromSecret: hf-token-secret
@@ -70,3 +135,6 @@ spec:
             - decode
             - --disaggregation-strategy
             - decode_first
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/trtllm/deploy/disagg_planner.yaml b/examples/backends/trtllm/deploy/disagg_planner.yaml
index 9324bd50896f..e022be59317e 100644
--- a/examples/backends/trtllm/deploy/disagg_planner.yaml
+++ b/examples/backends/trtllm/deploy/disagg_planner.yaml
@@ -9,6 +9,15 @@ spec:
   pvcs:
     - name: dynamo-pvc
       create: false
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://trtllm-disagg-planner-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: trtllm-disagg-planner
@@ -34,6 +43,58 @@ spec:
             - --router-temperature
             - "0.0"
             - --no-kv-events
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: trtllm-disagg-planner
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     Planner:
       dynamoNamespace: trtllm-disagg-planner
       envFromSecret: hf-token-secret
@@ -102,6 +163,9 @@ spec:
             - decode
             - --disaggregation-strategy
             - decode_first
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     TRTLLMPrefillWorker:
       dynamoNamespace: trtllm-disagg-planner
       envFromSecret: hf-token-secret
@@ -131,3 +195,6 @@ spec:
             - prefill
             - --disaggregation-strategy
             - decode_first
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/trtllm/deploy/disagg_router.yaml b/examples/backends/trtllm/deploy/disagg_router.yaml
index f687354a9546..2c4c3f9ff17f 100644
--- a/examples/backends/trtllm/deploy/disagg_router.yaml
+++ b/examples/backends/trtllm/deploy/disagg_router.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: trtllm-v1-disagg-router
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://trtllm-v1-disagg-router-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: trtllm-v1-disagg-router
@@ -17,6 +27,58 @@ spec:
       envs:
         - name: DYN_ROUTER_MODE
           value: kv
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: trtllm-v1-disagg-router
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     TRTLLMPrefillWorker:
       dynamoNamespace: trtllm-v1-disagg-router
       envFromSecret: hf-token-secret
@@ -45,6 +107,9 @@ spec:
             - --disaggregation-strategy
             - prefill_first
             - --publish-events-and-metrics
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     TRTLLMDecodeWorker:
       dynamoNamespace: trtllm-v1-disagg-router
       envFromSecret: hf-token-secret
@@ -72,3 +137,6 @@ spec:
             - decode
             - --disaggregation-strategy
             - prefill_first
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/trtllm/deploy/model_cache_pvc.yaml b/examples/backends/trtllm/deploy/model_cache_pvc.yaml
new file mode 100644
index 000000000000..7748488f4850
--- /dev/null
+++ b/examples/backends/trtllm/deploy/model_cache_pvc.yaml
@@ -0,0 +1,18 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: model-cache-pvc
+spec:
+  accessModes:
+    - ReadWriteMany
+  storageClassName: csi-mounted-fs-path-sc
+  resources:
+    requests:
+      storage: 256Gi
+    limits:
+      storage: 256Gi
+
+
diff --git a/examples/backends/vllm/deploy/README.md b/examples/backends/vllm/deploy/README.md
index 5f2c11dc4c38..28f4fa4d2c6a 100644
--- a/examples/backends/vllm/deploy/README.md
+++ b/examples/backends/vllm/deploy/README.md
@@ -9,6 +9,7 @@ Basic deployment pattern with frontend and a single decode worker.
 
 **Architecture:**
 - `Frontend`: OpenAI-compatible API server (with kv router mode disabled)
+- `ModelExpress`: Shared model caching service across workers
 - `VLLMDecodeWorker`: Single worker handling both prefill and decode
 
 ### 2. **Aggregated Router Deployment** (`agg_router.yaml`)
@@ -16,6 +17,7 @@ Enhanced aggregated deployment with KV cache routing capabilities.
 
 **Architecture:**
 - `Frontend`: OpenAI-compatible API server (with kv router mode enabled)
+- `ModelExpress`: Shared model caching service across workers
 - `VLLMDecodeWorker`: Single worker handling both prefill and decode
 
 ### 3. **Disaggregated Deployment** (`disagg.yaml`)
@@ -23,6 +25,7 @@ High-performance deployment with separated prefill and decode workers.
 
 **Architecture:**
 - `Frontend`: HTTP API server coordinating between workers
+- `ModelExpress`: Shared model caching service across workers
 - `VLLMDecodeWorker`: Specialized decode-only worker
 - `VLLMPrefillWorker`: Specialized prefill-only worker (`--is-prefill-worker`)
 - Communication via NIXL transfer backend
@@ -32,6 +35,7 @@ Advanced disaggregated deployment with KV cache routing capabilities.
 
 **Architecture:**
 - `Frontend`: HTTP API server with KV-aware routing
+- `ModelExpress`: Shared model caching service across workers
 - `VLLMDecodeWorker`: Specialized decode-only worker
 - `VLLMPrefillWorker`: Specialized prefill-only worker (`--is-prefill-worker`)
 
@@ -87,6 +91,18 @@ Before using these templates, ensure you have:
 3. **Container registry access** for vLLM runtime images
 4. **HuggingFace token secret** (referenced as `envFromSecret: hf-token-secret`)
 
+### Persistent Volume Claim (PVC)
+
+All templates expect a pre-created PVC named `model-cache-pvc` for the shared model cache used by ModelExpress and vLLM workers.
+
+Apply the shared PVC once per namespace before deploying any graph:
+
+```bash
+kubectl apply -f model_cache_pvc.yaml -n $NAMESPACE
+```
+
+Note: If your cluster requires a specific storage class, edit `model_cache_pvc.yaml` to set `storageClassName` accordingly.
+
 ### Container Images
 
 We have public images available on [NGC Catalog](https://catalog.ngc.nvidia.com/orgs/nvidia/teams/ai-dynamo/collections/ai-dynamo/artifacts). If you'd prefer to use your own registry, build and push your own image:
@@ -144,6 +160,9 @@ Export the NAMESPACE you used in your Dynamo Cloud Installation.
 cd <dynamo-source-root>/examples/backends/vllm/deploy
 export DEPLOYMENT_FILE=agg.yaml
 
+# Create the shared model cache PVC (run once per namespace)
+kubectl apply -f model_cache_pvc.yaml -n $NAMESPACE
+
 kubectl apply -f $DEPLOYMENT_FILE -n $NAMESPACE
 ```
 
diff --git a/examples/backends/vllm/deploy/agg.yaml b/examples/backends/vllm/deploy/agg.yaml
index 544fb7912415..81bee1308e44 100644
--- a/examples/backends/vllm/deploy/agg.yaml
+++ b/examples/backends/vllm/deploy/agg.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: vllm-agg
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://vllm-agg-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: vllm-agg
@@ -13,7 +23,59 @@ spec:
       replicas: 1
       extraPodSpec:
         mainContainer:
-          image: gitlab-master.nvidia.com:5005/dl/ai-dynamo/dynamo/dynamo:vllm-1029
+          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: vllm-agg
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmDecodeWorker:
       envFromSecret: hf-token-secret
       dynamoNamespace: vllm-agg
@@ -24,7 +86,7 @@ spec:
           gpu: "1"
       extraPodSpec:
         mainContainer:
-          image: gitlab-master.nvidia.com:5005/dl/ai-dynamo/dynamo/dynamo:vllm-1029
+          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
           workingDir: /workspace/examples/backends/vllm
           command:
             - python3
@@ -33,3 +95,6 @@ spec:
           args:
             - --model
             - Qwen/Qwen3-0.6B
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/vllm/deploy/agg_kvbm.yaml b/examples/backends/vllm/deploy/agg_kvbm.yaml
index 62e28386aa2a..24790aeda9ad 100644
--- a/examples/backends/vllm/deploy/agg_kvbm.yaml
+++ b/examples/backends/vllm/deploy/agg_kvbm.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: vllm-agg-kvbm
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://vllm-agg-kvbm-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: vllm-agg-kvbm
@@ -14,6 +24,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: vllm-agg-kvbm
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmDecodeWorker:
       envFromSecret: hf-token-secret
       dynamoNamespace: vllm-agg-kvbm
@@ -29,6 +91,7 @@ spec:
       envs:
         - name: DYN_KVBM_CPU_CACHE_GB
           value: "100"
+
       extraPodSpec:
         mainContainer:
           image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
@@ -48,3 +111,6 @@ spec:
             - --enforce-eager
             - --connector
             - kvbm
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/vllm/deploy/agg_router.yaml b/examples/backends/vllm/deploy/agg_router.yaml
index 26b961a06198..650a5baf7c1a 100644
--- a/examples/backends/vllm/deploy/agg_router.yaml
+++ b/examples/backends/vllm/deploy/agg_router.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: vllm-agg-router
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://vllm-agg-router-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: vllm-agg-router
@@ -17,6 +27,58 @@ spec:
       envs:
         - name: DYN_ROUTER_MODE
           value: kv
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: vllm-agg-router
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmDecodeWorker:
       envFromSecret: hf-token-secret
       dynamoNamespace: vllm-agg-router
@@ -36,3 +98,6 @@ spec:
           args:
             - --model
             - Qwen/Qwen3-0.6B
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/vllm/deploy/disagg-multinode.yaml b/examples/backends/vllm/deploy/disagg-multinode.yaml
index bfb56e6daf05..aaf1719d2576 100644
--- a/examples/backends/vllm/deploy/disagg-multinode.yaml
+++ b/examples/backends/vllm/deploy/disagg-multinode.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: vllm-disagg
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://vllm-disagg-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: vllm-disagg
@@ -22,6 +32,58 @@ spec:
           args:
             - --http-port
             - "8000"
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: vllm-disagg
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     decode:
       dynamoNamespace: vllm-disagg
       envFromSecret: hf-token-secret
@@ -45,6 +107,9 @@ spec:
             - Qwen/Qwen3-0.6B
             - --tensor-parallel-size
             - "2"
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     prefill:
       dynamoNamespace: vllm-disagg
       envFromSecret: hf-token-secret
@@ -69,3 +134,6 @@ spec:
             - --is-prefill-worker
             - --tensor-parallel-size
             - "2"
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/vllm/deploy/disagg.yaml b/examples/backends/vllm/deploy/disagg.yaml
index 799d8741b9d7..243e691736ea 100644
--- a/examples/backends/vllm/deploy/disagg.yaml
+++ b/examples/backends/vllm/deploy/disagg.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: vllm-disagg
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://vllm-disagg-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: vllm-disagg
@@ -14,6 +24,57 @@ spec:
       extraPodSpec:
         mainContainer:
           image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: vllm-disagg
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmDecodeWorker:
       dynamoNamespace: vllm-disagg
       envFromSecret: hf-token-secret
@@ -35,6 +96,9 @@ spec:
             - --model
             - Qwen/Qwen3-0.6B
             - --is-decode-worker
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmPrefillWorker:
       dynamoNamespace: vllm-disagg
       envFromSecret: hf-token-secret
@@ -56,3 +120,6 @@ spec:
             - --model
             - Qwen/Qwen3-0.6B
             - --is-prefill-worker
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/vllm/deploy/disagg_kvbm.yaml b/examples/backends/vllm/deploy/disagg_kvbm.yaml
index f4315a13cdd4..27dc0c122322 100644
--- a/examples/backends/vllm/deploy/disagg_kvbm.yaml
+++ b/examples/backends/vllm/deploy/disagg_kvbm.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: vllm-disagg-kvbm
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://vllm-disagg-kvbm-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: vllm-disagg-kvbm
@@ -14,6 +24,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: vllm-disagg-kvbm
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmDecodeWorker:
       dynamoNamespace: vllm-disagg-kvbm
       envFromSecret: hf-token-secret
@@ -39,6 +101,9 @@ spec:
             - --max-model-len
             - "32000"
             - --enforce-eager
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmPrefillWorker:
       dynamoNamespace: vllm-disagg-kvbm
       envFromSecret: hf-token-secret
@@ -75,3 +140,6 @@ spec:
             - --connector
             - kvbm
             - nixl
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/vllm/deploy/disagg_kvbm_2p2d.yaml b/examples/backends/vllm/deploy/disagg_kvbm_2p2d.yaml
index 1aa5281d09cf..48266742a316 100644
--- a/examples/backends/vllm/deploy/disagg_kvbm_2p2d.yaml
+++ b/examples/backends/vllm/deploy/disagg_kvbm_2p2d.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: vllm-disagg-kvbm-2p2d
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://vllm-disagg-kvbm-2p2d-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: vllm-disagg-kvbm-2p2d
@@ -14,6 +24,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: vllm-disagg-kvbm-2p2d
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmDecodeWorker:
       dynamoNamespace: vllm-disagg-kvbm-2p2d
       envFromSecret: hf-token-secret
@@ -39,6 +101,9 @@ spec:
             - --max-model-len
             - "32000"
             - --enforce-eager
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmPrefillWorker:
       dynamoNamespace: vllm-disagg-kvbm-2p2d
       envFromSecret: hf-token-secret
@@ -75,3 +140,6 @@ spec:
             - --connector
             - kvbm
             - nixl
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/vllm/deploy/disagg_kvbm_tp2.yaml b/examples/backends/vllm/deploy/disagg_kvbm_tp2.yaml
index 439b17a91fc0..d14d90faa70f 100644
--- a/examples/backends/vllm/deploy/disagg_kvbm_tp2.yaml
+++ b/examples/backends/vllm/deploy/disagg_kvbm_tp2.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: vllm-disagg-kvbm-tp2
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://vllm-disagg-kvbm-tp2-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: vllm-disagg-kvbm-tp2
@@ -14,6 +24,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: vllm-disagg-kvbm-tp2
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmDecodeWorker:
       dynamoNamespace: vllm-disagg-kvbm-tp2
       envFromSecret: hf-token-secret
@@ -43,6 +105,9 @@ spec:
             - --enforce-eager
             - --tensor-parallel-size
             - "2"
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmPrefillWorker:
       dynamoNamespace: vllm-disagg-kvbm-tp2
       envFromSecret: hf-token-secret
@@ -81,3 +146,6 @@ spec:
             - nixl
             - --tensor-parallel-size
             - "2"
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/vllm/deploy/disagg_planner.yaml b/examples/backends/vllm/deploy/disagg_planner.yaml
index 5e33a0d86634..5eaf6b137003 100644
--- a/examples/backends/vllm/deploy/disagg_planner.yaml
+++ b/examples/backends/vllm/deploy/disagg_planner.yaml
@@ -9,6 +9,15 @@ spec:
   pvcs:
     - name: dynamo-pvc
       create: false # Must be pre-created before deployment and SLA profiler must have been run
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://vllm-disagg-planner-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: vllm-disagg-planner
@@ -17,6 +26,58 @@ spec:
       extraPodSpec:
         mainContainer:
           image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: vllm-disagg-planner
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     Planner:
       dynamoNamespace: vllm-disagg-planner
       componentType: planner
@@ -57,6 +118,9 @@ spec:
             - dynamo.vllm
             - --model
             - Qwen/Qwen3-0.6B
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmPrefillWorker:
       dynamoNamespace: vllm-disagg-planner
       envFromSecret: hf-token-secret
@@ -78,3 +142,6 @@ spec:
             - --model
             - Qwen/Qwen3-0.6B
             - --is-prefill-worker
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/vllm/deploy/disagg_router.yaml b/examples/backends/vllm/deploy/disagg_router.yaml
index 0a13bace4944..fe3a92f055e9 100644
--- a/examples/backends/vllm/deploy/disagg_router.yaml
+++ b/examples/backends/vllm/deploy/disagg_router.yaml
@@ -6,6 +6,16 @@ kind: DynamoGraphDeployment
 metadata:
   name: vllm-v1-disagg-router
 spec:
+  pvcs:
+    - name: model-cache-pvc
+      create: false
+  envs:
+    - name: MODEL_EXPRESS_CACHE_DIRECTORY
+      value: "/model/.model-express/cache"
+    - name: HF_HUB_CACHE
+      value: "/model/.model-express/cache"
+    - name: MODEL_EXPRESS_URL
+      value: "http://vllm-v1-disagg-router-modelexpress:8000"
   services:
     Frontend:
       dynamoNamespace: vllm-v1-disagg-router
@@ -17,6 +27,58 @@ spec:
       envs:
         - name: DYN_ROUTER_MODE
           value: kv
+    ModelExpress:
+      envFromSecret: hf-token-secret
+      dynamoNamespace: vllm-v1-disagg-router
+      componentType: frontend
+      readinessProbe:
+        tcpSocket:
+          port: 8000
+        initialDelaySeconds: 10
+        periodSeconds: 5
+        timeoutSeconds: 3
+        failureThreshold: 3
+      replicas: 1
+      resources:
+        requests:
+          cpu: "4"
+          memory: "16Gi"
+        limits:
+          cpu: "4"
+          memory: "16Gi"
+      extraPodSpec:
+        mainContainer:
+          image: nvcr.io/nvidia/ai-dynamo/modelexpress-server:my-tag
+          imagePullPolicy: IfNotPresent
+          env:
+            - name: MODEL_EXPRESS_SERVER_PORT
+              value: "8000"
+            - name: MODEL_EXPRESS_LOGGING_LEVEL
+              value: "info"
+            - name: MODEL_EXPRESS_DATABASE_PATH
+              value: "/model/models.db"
+          command:
+            - /bin/sh
+            - -c
+          args:
+            - |
+              echo "Setting up Model Express configuration..."
+
+              mkdir -p $MODEL_EXPRESS_CACHE_DIRECTORY
+              cat > $MODEL_EXPRESS_CACHE_DIRECTORY/config.yaml << EOF
+              local_path: $MODEL_EXPRESS_CACHE_DIRECTORY
+              server_endpoint: http://localhost:8000
+              timeout_secs: null
+              EOF
+
+              ./modelexpress-server &
+
+              SERVER_PID=$!
+              echo "Server started with PID: $SERVER_PID"
+              wait $SERVER_PID
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmDecodeWorker:
       dynamoNamespace: vllm-v1-disagg-router
       envFromSecret: hf-token-secret
@@ -37,6 +99,9 @@ spec:
             - --model
             - Qwen/Qwen3-0.6B
             - --is-decode-worker
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
     VllmPrefillWorker:
       dynamoNamespace: vllm-v1-disagg-router
       envFromSecret: hf-token-secret
@@ -57,3 +122,6 @@ spec:
             - --model
             - Qwen/Qwen3-0.6B
             - --is-prefill-worker
+      volumeMounts:
+        - name: model-cache-pvc
+          mountPoint: /model
diff --git a/examples/backends/vllm/deploy/model_cache_pvc.yaml b/examples/backends/vllm/deploy/model_cache_pvc.yaml
new file mode 100644
index 000000000000..56ad5f005224
--- /dev/null
+++ b/examples/backends/vllm/deploy/model_cache_pvc.yaml
@@ -0,0 +1,16 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: model-cache-pvc
+spec:
+  accessModes:
+    - ReadWriteMany
+  storageClassName: csi-mounted-fs-path-sc
+  resources:
+    requests:
+      storage: 256Gi
+    limits:
+      storage: 256Gi