diff --git a/deploy/docker-compose.yml b/deploy/docker-compose.yml
index e02cdf0..bb25e44 100644
--- a/deploy/docker-compose.yml
+++ b/deploy/docker-compose.yml
@@ -19,8 +19,13 @@ services:
       dockerfile: deploy/Dockerfile
     image: prometheus:latest
     restart: unless-stopped
+    depends_on:
+      clamav:
+        condition: service_healthy
     environment:
       BNS_NETWORK: chipnet
+      CLAMD_HOST: clamav
+      CLAMD_PORT: "3310"
       ADMIN_PUBLISHER: ${ADMIN_PUBLISHER:-}
       VIRUSTOTAL_API_KEY: ${VIRUSTOTAL_API_KEY:-}
       PAYMENT_REQUIRED: ${PAYMENT_REQUIRED:-}
@@ -30,6 +35,7 @@ services:
     volumes:
       - data:/data
     networks:
+      - prometheus-internal
       - hephaestus
     expose:
       - "3700"
@@ -38,12 +44,39 @@ services:
       interval: 30s
       timeout: 5s
       retries: 3
-      start_period: 10s
+      start_period: 60s
+
+  # ClamAV runs as a sibling container on an internal network — only
+  # Prometheus can reach it. First start downloads ~250 MB of signatures via
+  # freshclam into a volume; subsequent starts reuse it.
+  clamav:
+    image: clamav/clamav:stable
+    restart: unless-stopped
+    environment:
+      CLAMAV_NO_MILTERD: "true"
+    volumes:
+      - clamav-sigs:/var/lib/clamav
+    networks:
+      - prometheus-internal
+    expose:
+      - "3310"
+    healthcheck:
+      test: ["CMD-SHELL", "clamdcheck.sh || clamdscan --ping 2>/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 10
+      # Signature download on first start can take a few minutes.
+      start_period: 5m
 
 volumes:
   data:
+  clamav-sigs:
 
 networks:
+  # Internal: only Prometheus + clamav. Clamd never exposed off this network.
+  prometheus-internal:
+    driver: bridge
+  # External: shared with Hephaestus so its Caddy can route prometheus.x.
   hephaestus:
     external: true
     name: hephaestus_default