diff --git a/truss/contexts/image_builder/serving_image_builder.py b/truss/contexts/image_builder/serving_image_builder.py
index 8e7fad9db..064fd57d1 100644
--- a/truss/contexts/image_builder/serving_image_builder.py
+++ b/truss/contexts/image_builder/serving_image_builder.py
@@ -378,17 +378,20 @@ def generate_docker_server_nginx_config(build_dir, config):
     nginx_filepath.write_text(nginx_content)
 
 
-def generate_docker_server_supervisord_config(build_dir, config):
-    supervisord_template = read_template_from_fs(
-        DOCKER_SERVER_TEMPLATES_DIR, "supervisord.conf.jinja"
+def generate_docker_server_wrapper_script(build_dir, config):
+    wrapper_template = read_template_from_fs(
+        DOCKER_SERVER_TEMPLATES_DIR, "server_wrapper.sh"
     )
     assert config.docker_server.start_command is not None, (
         "docker_server.start_command is required to use custom server"
     )
     start_command = config.docker_server.start_command
-    supervisord_contents = supervisord_template.render(start_command=start_command)
-    supervisord_filepath = build_dir / "supervisord.conf"
-    supervisord_filepath.write_text(supervisord_contents)
+    server_port = config.docker_server.server_port
+    wrapper_contents = wrapper_template.render(
+        start_command=start_command, server_port=server_port
+    )
+    wrapper_filepath = build_dir / "server_wrapper.sh"
+    wrapper_filepath.write_text(wrapper_contents)
 
 
 class ServingImageBuilderContext(TrussContext):
@@ -586,7 +589,7 @@ def prepare_image_build_dir(
 
             generate_docker_server_nginx_config(build_dir, config)
 
-            generate_docker_server_supervisord_config(build_dir, config)
+            generate_docker_server_wrapper_script(build_dir, config)
 
         # Override config.yml
         with (build_dir / CONFIG_FILE).open("w") as config_file:
diff --git a/truss/templates/docker_server/server_wrapper.sh b/truss/templates/docker_server/server_wrapper.sh
new file mode 100644
index 000000000..d83629034
--- /dev/null
+++ b/truss/templates/docker_server/server_wrapper.sh
@@ -0,0 +1,300 @@
+#!/bin/bash
+set -euo pipefail
+
+# Enhanced shell script to match supervisord behavior for custom servers
+# Manages nginx and model server processes with auto-restart and proper output handling
+
+# Global variables for process management
+declare -A PROCESS_PIDS
+declare -A RESTART_COUNTS
+declare -A LAST_RESTART_TIME
+declare -A PROCESS_START_TIME
+SHUTDOWN_REQUESTED=false
+
+# Configuration matching supervisord defaults
+MAX_RESTART_ATTEMPTS=3
+RESTART_RESET_TIME=10  # Reset restart counter after 10 seconds of stable operation
+FATAL_STATE_GRACE_PERIOD=5  # Wait 5 seconds before declaring fatal state
+LINEAR_BACKOFF_INTERVAL=1  # supervisord uses linear backoff by default
+
+log() {
+    echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*" >&2
+}
+
+# Function to start nginx with output redirection
+start_nginx() {
+    log "Starting nginx..."
+
+    # Start nginx with output redirected to stdout (like supervisord's stdout_logfile=/dev/fd/1)
+    nginx -g "daemon off;" &
+    local nginx_pid=$!
+
+    # Give nginx a moment to start
+    sleep 1
+
+    # Check if nginx started successfully
+    if kill -0 "$nginx_pid" 2>/dev/null; then
+        PROCESS_PIDS["nginx"]=$nginx_pid
+        PROCESS_START_TIME["nginx"]=$(date +%s)
+        log "Nginx started successfully (PID: $nginx_pid)"
+        return 0
+    else
+        log "ERROR: Failed to start nginx"
+        return 1
+    fi
+}
+
+# Function to start model server with output redirection
+start_model_server() {
+    log "Starting model server with command: $START_COMMAND"
+
+    # Start model server with output redirected to stdout (like supervisord)
+    # Using eval to handle complex commands properly
+    eval "$START_COMMAND" &
+    local model_pid=$!
+
+    # Wait for model server to be ready (similar to supervisord's startsecs=30)
+    log "Waiting for model server to be ready..."
+    for i in {1..30}; do
+        if ! kill -0 "$model_pid" 2>/dev/null; then
+            log "ERROR: Model server failed to start"
+            return 1
+        fi
+
+        # Check if server is responding (basic health check)
+        if curl -s -f "http://localhost:${SERVER_PORT}/ready" >/dev/null 2>&1; then
+            log "Model server is ready"
+            break
+        fi
+
+        if [[ $i -eq 30 ]]; then
+            log "WARNING: Model server readiness check timed out, continuing anyway"
+        fi
+
+        sleep 1
+    done
+
+    PROCESS_PIDS["model_server"]=$model_pid
+    PROCESS_START_TIME["model_server"]=$(date +%s)
+    log "Model server started successfully (PID: $model_pid)"
+    return 0
+}
+
+# Function to check if restart counter should be reset (like supervisord's startsecs behavior)
+should_reset_restart_counter() {
+    local process_name=$1
+    local current_time=$(date +%s)
+    local start_time=${PROCESS_START_TIME[$process_name]:-0}
+
+    # Reset counter if process has been running stably for RESTART_RESET_TIME seconds
+    if [[ $((current_time - start_time)) -gt $RESTART_RESET_TIME ]]; then
+        return 0
+    fi
+    return 1
+}
+
+# Function to restart a process with backoff (matching supervisord behavior)
+restart_process() {
+    local process_name=$1
+
+    # Check if we should reset the restart counter (process ran successfully for a while)
+    if should_reset_restart_counter "$process_name"; then
+        RESTART_COUNTS[$process_name]=0
+        log "Resetting restart counter for $process_name (process ran stably for $RESTART_RESET_TIME seconds)"
+    fi
+
+    local restart_count=${RESTART_COUNTS[$process_name]:-0}
+
+    if [[ $restart_count -ge $MAX_RESTART_ATTEMPTS ]]; then
+        log "ERROR: $process_name has reached max restart limit ($MAX_RESTART_ATTEMPTS)"
+        return 1
+    fi
+
+    # Use linear backoff like supervisord (not exponential)
+    local backoff_time=$((restart_count * LINEAR_BACKOFF_INTERVAL))
+
+    log "Restarting $process_name (attempt $((restart_count + 1))/$MAX_RESTART_ATTEMPTS) after ${backoff_time}s backoff"
+    sleep $backoff_time
+
+    # Increment restart count and record restart time
+    RESTART_COUNTS[$process_name]=$((restart_count + 1))
+    LAST_RESTART_TIME[$process_name]=$(date +%s)
+
+    # Restart the process
+    if [[ "$process_name" == "nginx" ]]; then
+        start_nginx
+    elif [[ "$process_name" == "model_server" ]]; then
+        start_model_server
+    fi
+
+    return $?
+}
+
+# Function to check if a process is running
+check_process() {
+    local process_name=$1
+    local pid=${PROCESS_PIDS[$process_name]:-}
+
+    if [[ -n "$pid" ]] && kill -0 "$pid" 2>/dev/null; then
+        return 0  # Process is running
+    else
+        return 1  # Process is not running
+    fi
+}
+
+# Function to handle process failures (implements supervisord's PROCESS_STATE_FATAL behavior)
+handle_process_failure() {
+    local process_name=$1
+
+    if [[ "$SHUTDOWN_REQUESTED" == "true" ]]; then
+        return 0  # Don't restart if shutdown was requested
+    fi
+
+    log "WARNING: $process_name process has stopped"
+
+    # Remove from process tracking
+    unset PROCESS_PIDS[$process_name]
+
+    # Attempt to restart the process
+    if restart_process "$process_name"; then
+        log "Successfully restarted $process_name"
+        return 0
+    else
+        log "ERROR: Failed to restart $process_name after max attempts"
+
+        # Implement supervisord's PROCESS_STATE_FATAL behavior - wait before declaring fatal
+        log "Waiting $FATAL_STATE_GRACE_PERIOD seconds before declaring fatal state..."
+        sleep $FATAL_STATE_GRACE_PERIOD
+
+        # Check if shutdown was requested during grace period
+        if [[ "$SHUTDOWN_REQUESTED" == "true" ]]; then
+            return 0
+        fi
+
+        log "ERROR: $process_name has entered FATAL state (exhausted restart attempts)"
+        return 1
+    fi
+}
+
+# Cleanup function for graceful shutdown
+cleanup() {
+    log "Received shutdown signal, stopping processes..."
+    SHUTDOWN_REQUESTED=true
+
+    # Stop model server if running
+    if check_process "model_server"; then
+        local model_pid=${PROCESS_PIDS["model_server"]}
+        log "Stopping model server (PID: $model_pid)"
+        kill -TERM "$model_pid" 2>/dev/null || true
+
+        # Wait for graceful shutdown (max 30 seconds)
+        for i in {1..30}; do
+            if ! kill -0 "$model_pid" 2>/dev/null; then
+                log "Model server stopped gracefully"
+                break
+            fi
+            if [[ $i -eq 30 ]]; then
+                log "Force killing model server"
+                kill -KILL "$model_pid" 2>/dev/null || true
+            fi
+            sleep 1
+        done
+    fi
+
+    # Stop nginx if running
+    if check_process "nginx"; then
+        local nginx_pid=${PROCESS_PIDS["nginx"]}
+        log "Stopping nginx (PID: $nginx_pid)"
+        kill -TERM "$nginx_pid" 2>/dev/null || true
+
+        # Wait for graceful shutdown (max 10 seconds)
+        for i in {1..10}; do
+            if ! kill -0 "$nginx_pid" 2>/dev/null; then
+                log "Nginx stopped gracefully"
+                break
+            fi
+            if [[ $i -eq 10 ]]; then
+                log "Force killing nginx"
+                kill -KILL "$nginx_pid" 2>/dev/null || true
+            fi
+            sleep 1
+        done
+    fi
+
+    log "Shutdown complete"
+    exit 0
+}
+
+# Set up signal handlers
+trap cleanup SIGTERM SIGINT
+
+# Validate required environment variables
+if [[ -z "${START_COMMAND:-}" ]]; then
+    log "ERROR: START_COMMAND environment variable is required"
+    exit 1
+fi
+
+if [[ -z "${SERVER_PORT:-}" ]]; then
+    log "ERROR: SERVER_PORT environment variable is required"
+    exit 1
+fi
+
+log "Starting custom server wrapper"
+log "Model server command: $START_COMMAND"
+log "Server port: $SERVER_PORT"
+
+# Initialize restart counts using associative arrays (properly scoped)
+RESTART_COUNTS["nginx"]=0
+RESTART_COUNTS["model_server"]=0
+
+# Start both processes
+if ! start_nginx; then
+    log "ERROR: Failed to start nginx initially"
+    exit 1
+fi
+
+if ! start_model_server; then
+    log "ERROR: Failed to start model server initially"
+    cleanup
+    exit 1
+fi
+
+log "Both services are running, monitoring processes..."
+
+# Main monitoring loop - matches supervisord's behavior
+FATAL_STATE_REACHED=false
+
+while [[ "$FATAL_STATE_REACHED" == "false" && "$SHUTDOWN_REQUESTED" == "false" ]]; do
+    # Check nginx status
+    if ! check_process "nginx"; then
+        log "WARNING: Nginx process has stopped"
+        if ! handle_process_failure "nginx"; then
+            log "ERROR: Nginx has entered FATAL state"
+            FATAL_STATE_REACHED=true
+        fi
+    fi
+
+    # Check model server status (only if we haven't reached fatal state)
+    if [[ "$FATAL_STATE_REACHED" == "false" ]] && ! check_process "model_server"; then
+        log "WARNING: Model server process has stopped"
+        if ! handle_process_failure "model_server"; then
+            log "ERROR: Model server has entered FATAL state"
+            FATAL_STATE_REACHED=true
+        fi
+    fi
+
+    # Sleep for a short interval before checking again (like supervisord)
+    sleep 5
+done
+
+# Handle fatal state (like supervisord's PROCESS_STATE_FATAL)
+if [[ "$FATAL_STATE_REACHED" == "true" ]]; then
+    log "ERROR: One or more processes have entered FATAL state - shutting down"
+    cleanup
+    exit 1
+fi
+
+# Normal shutdown
+if [[ "$SHUTDOWN_REQUESTED" == "true" ]]; then
+    cleanup
+fi
diff --git a/truss/templates/docker_server_requirements.txt b/truss/templates/docker_server_requirements.txt
index 3131eb911..7e31bd68d 100644
--- a/truss/templates/docker_server_requirements.txt
+++ b/truss/templates/docker_server_requirements.txt
@@ -1 +1 @@
-supervisor==4.2.5
+# Shell script wrapper replaces supervisord for process management
diff --git a/truss/templates/server.Dockerfile.jinja b/truss/templates/server.Dockerfile.jinja
index cbe06cd51..dfeeaf4e5 100644
--- a/truss/templates/server.Dockerfile.jinja
+++ b/truss/templates/server.Dockerfile.jinja
@@ -119,17 +119,18 @@ RUN uv python install {{ control_python_version }}
 RUN uv venv /docker_server/.venv --python {{ control_python_version }}
 RUN uv pip install --python /docker_server/.venv/bin/python -r /app/docker_server_requirements.txt --no-cache-dir
 {% set proxy_config_path = "/etc/nginx/conf.d/proxy.conf" %}
-{% set supervisor_config_path = "/etc/supervisor/supervisord.conf" %}
-{% set supervisor_server_url = "http://localhost:8080" %}
+{% set server_wrapper_path = "/docker_server/server_wrapper.sh" %}
 COPY --chown={{ default_owner }} ./proxy.conf {{ proxy_config_path }}
-COPY --chown={{ default_owner }} ./supervisord.conf {{ supervisor_config_path }}
-ENV SUPERVISOR_SERVER_URL="{{ supervisor_server_url }}"
-ENV SERVER_START_CMD="/docker_server/.venv/bin/supervisord -c {{ supervisor_config_path }}"
+COPY --chown={{ default_owner }} ./server_wrapper.sh {{ server_wrapper_path }}
+RUN chmod +x {{ server_wrapper_path }}
+ENV START_COMMAND={{ config.docker_server.start_command | tojson }}
+ENV SERVER_PORT={{ config.docker_server.server_port | tojson }}
+ENV SERVER_START_CMD={{ server_wrapper_path | tojson }}
 {#- default configuration uses port 80, which requires root privileges, so we remove it #}
 RUN rm -f /etc/nginx/sites-enabled/default
 {#- nginx writes to /var/lib/nginx, /var/log/nginx, and /run directories #}
 {{ chown_and_switch_to_regular_user_if_enabled(["/var/lib/nginx", "/var/log/nginx", "/run"]) }}
-ENTRYPOINT ["/docker_server/.venv/bin/supervisord", "-c", "{{ supervisor_config_path }}"]
+ENTRYPOINT ["{{ server_wrapper_path }}"]
 
     {%- elif requires_live_reload %} {#- elif requires_live_reload #}
 ENV HASH_TRUSS="{{ truss_hash }}"