diff --git a/docker/entrypoint.sh b/docker/entrypoint.sh
index f38334a8fe7..2974c895e08 100755
--- a/docker/entrypoint.sh
+++ b/docker/entrypoint.sh
@@ -9,10 +9,19 @@ if [ -d "/app/.cache/prisma-python/binaries" ] && [ ! -d "/tmp/.cache/prisma-pyt
     echo "Prisma binaries copied successfully"
 fi
 
-# Set runtime cache locations to /tmp (writable even in read-only root filesystem)
-export XDG_CACHE_HOME="${XDG_CACHE_HOME:-/tmp/.cache}"
-export NPM_CONFIG_CACHE="${NPM_CONFIG_CACHE:-/tmp/npm-cache}"
-export PRISMA_BINARY_CACHE_DIR="${PRISMA_BINARY_CACHE_DIR:-/tmp/.cache/prisma-python/binaries}"
+# Override Dockerfile ENV defaults that point to read-only /app/.cache,
+# but respect user-provided overrides (e.g. via K8s pod spec, docker-compose).
+# The Dockerfile.non_root sets these to /app/.cache/... at build time, but at
+# runtime /app is read-only, so we redirect to /tmp which is always writable.
+if [ "$XDG_CACHE_HOME" = "/app/.cache" ] || [ -z "$XDG_CACHE_HOME" ]; then
+    export XDG_CACHE_HOME="/tmp/.cache"
+fi
+if [ "$NPM_CONFIG_CACHE" = "/app/.cache/npm" ] || [ -z "$NPM_CONFIG_CACHE" ]; then
+    export NPM_CONFIG_CACHE="/tmp/npm-cache"
+fi
+if [ "$PRISMA_BINARY_CACHE_DIR" = "/app/.cache/prisma-python/binaries" ] || [ -z "$PRISMA_BINARY_CACHE_DIR" ]; then
+    export PRISMA_BINARY_CACHE_DIR="/tmp/.cache/prisma-python/binaries"
+fi
 
 # Run the Python migration script
 python3 litellm/proxy/prisma_migration.py
diff --git a/docker/prod_entrypoint.sh b/docker/prod_entrypoint.sh
index 7c6eea6b9fb..18ae4a4bce2 100644
--- a/docker/prod_entrypoint.sh
+++ b/docker/prod_entrypoint.sh
@@ -8,10 +8,19 @@ if [ -d "/app/.cache/prisma-python/binaries" ] && [ ! -d "/tmp/.cache/prisma-pyt
     echo "Prisma binaries copied successfully"
 fi
 
-# Set runtime cache locations to /tmp (writable even in read-only root filesystem)
-export XDG_CACHE_HOME="${XDG_CACHE_HOME:-/tmp/.cache}"
-export NPM_CONFIG_CACHE="${NPM_CONFIG_CACHE:-/tmp/npm-cache}"
-export PRISMA_BINARY_CACHE_DIR="${PRISMA_BINARY_CACHE_DIR:-/tmp/.cache/prisma-python/binaries}"
+# Override Dockerfile ENV defaults that point to read-only /app/.cache,
+# but respect user-provided overrides (e.g. via K8s pod spec, docker-compose).
+# The Dockerfile.non_root sets these to /app/.cache/... at build time, but at
+# runtime /app is read-only, so we redirect to /tmp which is always writable.
+if [ "$XDG_CACHE_HOME" = "/app/.cache" ] || [ -z "$XDG_CACHE_HOME" ]; then
+    export XDG_CACHE_HOME="/tmp/.cache"
+fi
+if [ "$NPM_CONFIG_CACHE" = "/app/.cache/npm" ] || [ -z "$NPM_CONFIG_CACHE" ]; then
+    export NPM_CONFIG_CACHE="/tmp/npm-cache"
+fi
+if [ "$PRISMA_BINARY_CACHE_DIR" = "/app/.cache/prisma-python/binaries" ] || [ -z "$PRISMA_BINARY_CACHE_DIR" ]; then
+    export PRISMA_BINARY_CACHE_DIR="/tmp/.cache/prisma-python/binaries"
+fi
 
 if [ "$SEPARATE_HEALTH_APP" = "1" ]; then
     export LITELLM_ARGS="$@"