diff --git a/.env.example b/.env.example
index 88ab36e..61c2885 100644
--- a/.env.example
+++ b/.env.example
@@ -3,6 +3,11 @@
 #
 #   cp .env.example .env
 
+# Address the dev ports are published on. 127.0.0.1 keeps them off the LAN.
+# Use 0.0.0.0 only if your hosts file maps host.docker.internal to the LAN IP
+# (then the browser can't reach 127.0.0.1-only ports through that name).
+DEV_BIND_ADDRESS=127.0.0.1
+
 # --- PostgreSQL (shared instance, one database per project) -----------------
 POSTGRES_USER=postgres
 POSTGRES_PASSWORD=postgres
diff --git a/README.md b/README.md
index 7d132da..c73a447 100644
--- a/README.md
+++ b/README.md
@@ -38,7 +38,9 @@ docker compose -f compose.yml -f compose.dev.yml up -d
 ```
 
 That starts Postgres, Zitadel (and its login UI), SeaweedFS, RabbitMQ and Jaeger, with their
-ports published to `localhost`. Add `--profile tools` to also start pgAdmin.
+ports published on `127.0.0.1` only (`DEV_BIND_ADDRESS`), so the dev
+credentials aren't reachable from the LAN. Add `--profile tools` to also
+start pgAdmin.
 
 In production, the base `compose.yml` file **is** the production shape
 already. There are no published ports, and everything is reachable only on
@@ -106,9 +108,8 @@ match this repo's `.env`". This file is the source of truth for them.
 
 ### The Zitadel admin PAT
 
-`zitadel-init` (in the compose file of whichever project runs it first)
-writes a service-account personal access token to `zitadel/.output/admin-sa.pat`
-on first init. Every consuming project mounts that same folder read-only, via
+Zitadel itself writes a service-account personal access token to
+`zitadel/.output/admin-sa.pat` when the instance is created. Every consuming project mounts that same folder read-only, via
 its own `ZITADEL_ADMIN_PAT_DIR` setting, to call the Zitadel management API
 and create its own project/apps.
 
@@ -180,7 +181,9 @@ docker compose -f edge/compose.yml up -d
   for new hostnames (their DNS has to already point at the host):
 
   ```bash
-  docker compose -f edge/compose.yml run --rm --entrypoint certbot certbot     certonly --webroot -w /var/www/certbot -d host1.example.com -d host2.example.com     --non-interactive --agree-tos -m you@example.com
+  docker compose -f edge/compose.yml run --rm --entrypoint certbot certbot \
+    certonly --webroot -w /var/www/certbot -d host1.example.com -d host2.example.com \
+    --non-interactive --agree-tos -m you@example.com
   ```
 
   Then add the `server` blocks pointing at
diff --git a/compose.dev.yml b/compose.dev.yml
index b235237..f62ab4f 100644
--- a/compose.dev.yml
+++ b/compose.dev.yml
@@ -5,43 +5,49 @@ name: infra-dev
 # of joining infra-net. Adds pgAdmin.
 #
 #   docker compose -f compose.yml -f compose.dev.yml up -d
+#
+# Ports are bound to DEV_BIND_ADDRESS (127.0.0.1 by default), so the dev
+# credentials in .env are not reachable from the LAN. Other containers still
+# reach them through host.docker.internal. The browser on the host does too
+# as long as the hosts file maps host.docker.internal to 127.0.0.1; if yours
+# maps it to the LAN IP, set DEV_BIND_ADDRESS=0.0.0.0 (see .env.example).
 
 services:
   postgres:
     ports:
-      - "${POSTGRES_HOST_PORT}:5432"
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:${POSTGRES_HOST_PORT}:5432"
 
   zitadel:
     ports:
-      - "${ZITADEL_EXTERNAL_PORT}:8080"
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:${ZITADEL_EXTERNAL_PORT}:8080"
 
   # No edge in dev to put the login under Zitadel's own port, so it gets its
   # own (ZITADEL_LOGIN_BASE_URL in .env points the browser at it).
   zitadel-login:
     ports:
-      - "${ZITADEL_LOGIN_HOST_PORT}:3000"
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:${ZITADEL_LOGIN_HOST_PORT}:3000"
 
   seaweedfs:
     ports:
-      - "9333:9333" # master / admin UI
-      - "8888:8888" # filer
-      - "8333:8333" # S3 API
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:9333:9333" # master / admin UI
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:8888:8888" # filer
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:8333:8333" # S3 API
 
   rabbitmq:
     ports:
-      - "5672:5672"   # AMQP
-      - "15672:15672" # management UI
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:5672:5672"   # AMQP
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:15672:15672" # management UI
 
   jaeger:
     ports:
-      - "16686:16686" # UI
-      - "4317:4317"   # OTLP gRPC
-      - "4318:4318"   # OTLP HTTP
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:16686:16686" # UI
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:4317:4317"   # OTLP gRPC
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:4318:4318"   # OTLP HTTP
 
   # Optional Postgres UI: docker compose ... --profile tools up -d
   pgadmin:
     profiles: [tools]
-    image: dpage/pgadmin4:latest
+    image: dpage/pgadmin4:9.18
     container_name: infra-pgadmin
     environment:
       PGADMIN_DEFAULT_EMAIL: ${PGADMIN_DEFAULT_EMAIL}
@@ -51,9 +57,15 @@ services:
       postgres:
         condition: service_healthy
     ports:
-      - "${PGADMIN_HOST_PORT}:80"
+      - "${DEV_BIND_ADDRESS:-127.0.0.1}:${PGADMIN_HOST_PORT}:80"
     volumes:
       - pgadmin_data:/var/lib/pgadmin
+    # Same rotation as compose.yml (YAML anchors don't cross files).
+    logging:
+      driver: json-file
+      options:
+        max-size: "10m"
+        max-file: "3"
     restart: unless-stopped
 
 volumes:
diff --git a/compose.yml b/compose.yml
index d7964e6..eeaf141 100644
--- a/compose.yml
+++ b/compose.yml
@@ -7,13 +7,21 @@
 # of joining a network. See README.md for why.
 #
 #   dev   docker compose -f compose.yml -f compose.dev.yml up -d
-#   prod  docker compose --env-file .env.prod -f compose.yml -f compose.prod.yml up -d
+#   prod  docker compose --env-file .env.prod -f compose.yml up -d
 #
 # A consuming project's own deploy stack does NOT get its own copy of any of
 # these services. It only needs this stack to already be up, and its own
 # one-shot bootstrap jobs (zitadel-init, seaweedfs-init) to create its own
 # OIDC apps / bucket against the shared instances below.
 
+# Docker's default json-file driver never rotates, so logs would grow until
+# the disk fills up.
+x-logging: &default-logging
+  driver: json-file
+  options:
+    max-size: "10m"
+    max-file: "3"
+
 networks:
   infra-net:
     name: infra-net
@@ -38,6 +46,7 @@ services:
       timeout: 5s
       retries: 10
     networks: [infra-net]
+    logging: *default-logging
     restart: unless-stopped
 
   # --- Identity -----------------------------------------------------------
@@ -118,6 +127,7 @@ services:
       retries: 12
       start_period: 20s
     networks: [infra-net]
+    logging: *default-logging
     restart: unless-stopped
 
   # Login V2 UI (Next.js), served under /ui/v2/login. It calls the API
@@ -143,6 +153,7 @@ services:
     volumes:
       - zitadel_bootstrap:/bootstrap:ro
     networks: [infra-net]
+    logging: *default-logging
     restart: unless-stopped
 
   # --- Storage --------------------------------------------------------------
@@ -152,7 +163,7 @@ services:
   # handed to consuming projects. No bucket is created here; each
   # consuming project's own seaweedfs-init one-shot creates its own.
   seaweedfs:
-    image: chrislusf/seaweedfs:latest
+    image: chrislusf/seaweedfs:4.47
     container_name: infra-seaweedfs
     entrypoint: ["/bin/sh", "-c"]
     command:
@@ -184,6 +195,7 @@ services:
       timeout: 5s
       retries: 10
     networks: [infra-net]
+    logging: *default-logging
     restart: unless-stopped
 
   # --- Messaging --------------------------------------------------------
@@ -207,16 +219,28 @@ services:
       timeout: 5s
       retries: 10
     networks: [infra-net]
+    logging: *default-logging
     restart: unless-stopped
 
   # --- Tracing --------------------------------------------------------------
   # OTLP collector + trace UI, shared by any project's OpenTelemetry exporter.
+  # Last Jaeger v1 release (v1 is deprecated; v2 is jaegertracing/jaeger).
   jaeger:
-    image: jaegertracing/all-in-one:latest
+    image: jaegertracing/all-in-one:1.76.0
     container_name: infra-jaeger
     environment:
       COLLECTOR_OTLP_ENABLED: "true"
+      # Traces are kept in memory; without a cap they grow until the
+      # container runs out of RAM. The oldest are dropped first.
+      MEMORY_MAX_TRACES: "100000"
+    healthcheck:
+      # Admin port: answers 200 once the collector and query service are up.
+      test: ["CMD", "wget", "-q", "-O", "-", "http://localhost:14269/"]
+      interval: 10s
+      timeout: 5s
+      retries: 10
     networks: [infra-net]
+    logging: *default-logging
     restart: unless-stopped
 
 volumes:
diff --git a/edge/compose.yml b/edge/compose.yml
index 9199f3d..4e882f1 100644
--- a/edge/compose.yml
+++ b/edge/compose.yml
@@ -10,6 +10,14 @@
 
 name: edge
 
+# Docker's default json-file driver never rotates, so logs would grow until
+# the disk fills up.
+x-logging: &default-logging
+  driver: json-file
+  options:
+    max-size: "10m"
+    max-file: "3"
+
 networks:
   infra-net:
     name: infra-net
@@ -36,10 +44,11 @@ services:
       - ./certbot/www:/var/www/certbot:ro
       - ./certbot/conf:/etc/letsencrypt:ro
     networks: [gitea, infra-net]
+    logging: *default-logging
     restart: unless-stopped
 
   certbot:
-    image: certbot/certbot
+    image: certbot/certbot:v5.8.0
     container_name: edge-certbot
     entrypoint:
       - /bin/sh
@@ -48,4 +57,5 @@ services:
     volumes:
       - ./certbot/conf:/etc/letsencrypt
       - ./certbot/www:/var/www/certbot
+    logging: *default-logging
     restart: unless-stopped