Always-On Agent Design Patterns
Background processes, daemon agents, and persistent connection management.
Always-On Agent Design Patterns is a free AI Agents lesson on CoddyKit — lesson 1 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the AI Agents learning path, one of 4 lessons in the course, and your progress syncs across the web and the CoddyKit app.
What Is an Always-On Agent?
An always-on agent runs continuously as a background service, waiting for events and taking action proactively. Unlike request-response agents, it persists between interactions and maintains state over time.
Daemon Process Pattern
A daemon process runs in the background, independent of any terminal session. Use Python's daemon threads or a process supervisor to keep the agent running after the terminal closes.
import threading
import time
import signal
import sys
shutdown_flag = threading.Event()
def agent_main_loop():
print('Agent daemon started')
while not shutdown_flag.is_set():
try:
# Agent work: check for events, process tasks
perform_agent_cycle()
shutdown_flag.wait(timeout=60) # Sleep 60s, wakes on shutdown
except Exception as e:
print(f'Agent loop error: {e}')
shutdown_flag.wait(timeout=5) # Brief pause on error
print('Agent daemon stopped')
def perform_agent_cycle():
print(f'Agent cycle at {time.strftime("%H:%M:%S")}')
# Check emails, process queue, run scheduled tasks
def handle_signal(signum, frame):
print(f'Signal {signum} received, shutting down...')
shutdown_flag.set()
# Register signal handlers for graceful shutdown
signal.signal(signal.SIGTERM, handle_signal)
signal.signal(signal.SIGINT, handle_signal)
# Start as daemon thread
thread = threading.Thread(target=agent_main_loop, daemon=True)
thread.start()
print('Agent running in background')Watchdog: Auto-Restart on Crash
A watchdog monitors the agent process and restarts it if it crashes. This is essential for production: agents inevitably hit unexpected errors and must recover automatically.
import subprocess
import time
import logging
logger = logging.getLogger('watchdog')
class AgentWatchdog:
def __init__(self, agent_script: str, max_restarts: int = 10, restart_delay: float = 5.0):
self.agent_script = agent_script
self.max_restarts = max_restarts
self.restart_delay = restart_delay
self.restart_count = 0
self.process = None
def start(self):
self.process = subprocess.Popen(
['python', self.agent_script],
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT
)
logger.info(f'Agent started (PID: {self.process.pid})')
def run_forever(self):
self.start()
while True:
return_code = self.process.wait()
logger.warning(f'Agent exited with code {return_code}')
if self.restart_count >= self.max_restarts:
logger.error(f'Max restarts ({self.max_restarts}) reached. Stopping watchdog.')
break
self.restart_count += 1
logger.info(f'Restarting agent (attempt {self.restart_count})...')
time.sleep(self.restart_delay * self.restart_count) # Backoff
self.start()
print('AgentWatchdog defined')Persistent WebSocket Connection
For real-time event delivery, maintain a persistent WebSocket connection. Reconnect automatically if the connection drops — this is the key challenge with always-on agents.
import asyncio
import websockets
import json
async def persistent_websocket_connection(uri: str, on_message):
backoff = 1
max_backoff = 60
while True: # Reconnect forever
try:
print(f'Connecting to {uri}')
async with websockets.connect(uri, ping_interval=30, ping_timeout=10) as ws:
print('WebSocket connected')
backoff = 1 # Reset backoff on successful connection
async for raw_message in ws:
try:
message = json.loads(raw_message)
await on_message(message)
except json.JSONDecodeError:
print(f'Invalid JSON message: {raw_message[:100]}')
except websockets.exceptions.ConnectionClosed as e:
print(f'WebSocket closed: {e}. Reconnecting in {backoff}s')
except Exception as e:
print(f'WebSocket error: {e}. Reconnecting in {backoff}s')
await asyncio.sleep(backoff)
backoff = min(backoff * 2, max_backoff) # Exponential backoff
async def handle_ws_message(message: dict):
print(f'Received: {message}')
print('Persistent WebSocket connection function defined')Heartbeat Checks
A heartbeat confirms the agent is alive and processing. Send a heartbeat signal every N seconds; if heartbeats stop, the watchdog knows the agent is stuck or dead.
import threading
import time
from datetime import datetime
class HeartbeatMonitor:
def __init__(self, max_silence_seconds: int = 300):
self.last_heartbeat = datetime.utcnow()
self.max_silence = max_silence_seconds
self.lock = threading.Lock()
def beat(self):
with self.lock:
self.last_heartbeat = datetime.utcnow()
def is_alive(self) -> bool:
with self.lock:
silence = (datetime.utcnow() - self.last_heartbeat).total_seconds()
return silence < self.max_silence
def silence_seconds(self) -> float:
with self.lock:
return (datetime.utcnow() - self.last_heartbeat).total_seconds()
monitor = HeartbeatMonitor(max_silence_seconds=60)
def agent_with_heartbeat():
while not shutdown_flag.is_set():
# Send heartbeat at start of each cycle
monitor.beat()
# Do agent work
perform_agent_cycle()
time.sleep(30)
# External watchdog checks the monitor
def watchdog_check():
while True:
if not monitor.is_alive():
print(f'ALERT: Agent silent for {monitor.silence_seconds():.0f}s')
# Restart agent here
time.sleep(30)
print('Heartbeat monitor defined')Graceful Shutdown
Graceful shutdown completes in-progress work before stopping. It receives a stop signal, prevents new work from starting, finishes current tasks, saves state, and exits cleanly.
import signal
import threading
from contextlib import contextmanager
class GracefulShutdown:
def __init__(self, timeout: float = 30.0):
self.should_stop = threading.Event()
self.active_tasks = 0
self.lock = threading.Lock()
self.timeout = timeout
signal.signal(signal.SIGTERM, self._handle_signal)
signal.signal(signal.SIGINT, self._handle_signal)
def _handle_signal(self, signum, frame):
print(f'Shutdown signal received. Waiting for {self.active_tasks} active tasks...')
self.should_stop.set()
@contextmanager
def task(self):
if self.should_stop.is_set():
raise RuntimeError('Shutdown in progress, not accepting new tasks')
with self.lock:
self.active_tasks += 1
try:
yield
finally:
with self.lock:
self.active_tasks -= 1
def wait_for_all_tasks(self):
self.should_stop.wait()
deadline = time.time() + self.timeout
while self.active_tasks > 0 and time.time() < deadline:
time.sleep(0.1)
if self.active_tasks > 0:
print(f'WARNING: Forced shutdown with {self.active_tasks} tasks still active')
shutdown = GracefulShutdown(timeout=30)
print('Graceful shutdown manager created')Handling Disconnects and Reconnects
Always-on agents need strategies for handling service disconnections: buffer events during downtime, replay missed events after reconnect, and avoid losing events that arrived while disconnected.
import asyncio
import redis
from datetime import datetime
r = redis.Redis(host='localhost', port=6379, decode_responses=True)
class DisconnectHandler:
def __init__(self, buffer_key: str = 'agent:offline_buffer'):
self.buffer_key = buffer_key
self.connected = True
def on_disconnect(self):
self.connected = False
print(f'Disconnected at {datetime.utcnow()}')
def on_reconnect(self):
self.connected = True
print(f'Reconnected at {datetime.utcnow()}')
self.replay_buffered_events()
def handle_event(self, event: dict):
if not self.connected:
# Buffer events for later replay
import json
r.lpush(self.buffer_key, json.dumps(event))
print(f'Event buffered (offline): {event["type"]}')
return
self.process_event(event)
def replay_buffered_events(self):
import json
replayed = 0
while True:
raw = r.rpop(self.buffer_key)
if not raw:
break
event = json.loads(raw)
self.process_event(event)
replayed += 1
if replayed:
print(f'Replayed {replayed} buffered events')
def process_event(self, event: dict):
print(f'Processing event: {event["type"]}')
handler = DisconnectHandler()
print('Disconnect handler created')Process Supervision with systemd
For Linux production deployments, use systemd to manage the agent process. It handles auto-restart, logging to journald, and starts the agent on boot.
# /etc/systemd/system/my-agent.service
SERVICE_FILE = '''
[Unit]
Description=My AI Agent Service
After=network.target
[Service]
Type=simple
User=ubuntu
WorkingDirectory=/home/ubuntu/agent
ExecStart=/home/ubuntu/venv/bin/python agent.py
Restart=always
RestartSec=10
StandardOutput=journal
StandardError=journal
# Environment variables
EnvironmentFile=/home/ubuntu/agent/.env
# Resource limits
MemoryLimit=1G
CPUQuota=50%
[Install]
WantedBy=multi-user.target
'''
# Deploy commands:
# sudo cp my-agent.service /etc/systemd/system/
# sudo systemctl daemon-reload
# sudo systemctl enable my-agent
# sudo systemctl start my-agent
# sudo systemctl status my-agent
# sudo journalctl -u my-agent -f # Follow logs
print('Systemd service configuration defined')
print('Enables: auto-start on boot, auto-restart on crash, centralized logging')State Persistence Across Restarts
An always-on agent must save its state so it can resume where it left off after a restart. Save checkpoint data to disk or Redis at regular intervals and after significant state changes.
import json
import os
from datetime import datetime
CHECKPOINT_FILE = '/tmp/agent_checkpoint.json'
def save_checkpoint(state: dict):
state['last_saved'] = datetime.utcnow().isoformat()
with open(CHECKPOINT_FILE, 'w') as f:
json.dump(state, f, indent=2)
print(f'Checkpoint saved at {state["last_saved"]}')
def load_checkpoint() -> dict:
if not os.path.exists(CHECKPOINT_FILE):
print('No checkpoint found, starting fresh')
return {}
with open(CHECKPOINT_FILE) as f:
state = json.load(f)
print(f'Checkpoint loaded from {state.get("last_saved", "unknown")}')
return state
# Agent startup
agent_state = load_checkpoint()
last_processed_id = agent_state.get('last_processed_email_id', 0)
print(f'Resuming from email ID: {last_processed_id}')
# After processing each email
agent_state['last_processed_email_id'] = last_processed_id + 1
if agent_state['last_processed_email_id'] % 10 == 0: # Checkpoint every 10 items
save_checkpoint(agent_state)Monitoring Always-On Agents
Track key metrics for always-on agents: uptime, events processed per hour, error rate, memory usage, and last active time. Expose these via a health endpoint or push to a monitoring service.
from fastapi import FastAPI
from datetime import datetime
import psutil
import os
app = FastAPI()
start_time = datetime.utcnow()
events_processed = 0
last_event_time = None
@app.get('/health')
def health_check():
process = psutil.Process(os.getpid())
uptime_seconds = (datetime.utcnow() - start_time).total_seconds()
last_active = None
if last_event_time:
last_active = (datetime.utcnow() - last_event_time).total_seconds()
return {
'status': 'ok',
'uptime_seconds': round(uptime_seconds),
'events_processed': events_processed,
'memory_mb': round(process.memory_info().rss / 1024 / 1024, 1),
'cpu_percent': process.cpu_percent(interval=1),
'last_event_seconds_ago': round(last_active) if last_active else None,
'timestamp': datetime.utcnow().isoformat()
}Circuit Breaker for External Dependencies
An always-on agent interacts with external services that can fail. A circuit breaker stops making calls to a failing service for a period, preventing cascade failures.
import time
from enum import Enum
class CircuitState(Enum):
CLOSED = 'closed' # Normal operation
OPEN = 'open' # Service down, not calling
HALF_OPEN = 'half_open' # Testing if service recovered
class CircuitBreaker:
def __init__(self, failure_threshold=5, recovery_timeout=60):
self.state = CircuitState.CLOSED
self.failure_count = 0
self.threshold = failure_threshold
self.recovery_timeout = recovery_timeout
self.last_failure_time = None
def call(self, fn, *args):
if self.state == CircuitState.OPEN:
if time.time() - self.last_failure_time > self.recovery_timeout:
self.state = CircuitState.HALF_OPEN
else:
raise RuntimeError('Circuit open: service unavailable')
try:
result = fn(*args)
if self.state == CircuitState.HALF_OPEN:
self.state = CircuitState.CLOSED
self.failure_count = 0
print('Circuit closed: service recovered')
return result
except Exception as e:
self.failure_count += 1
self.last_failure_time = time.time()
if self.failure_count >= self.threshold:
self.state = CircuitState.OPEN
print(f'Circuit opened after {self.failure_count} failures')
raise
circuit = CircuitBreaker(failure_threshold=3, recovery_timeout=30)
print('Circuit breaker created')Knowledge Check: Always-On Agents
Test your understanding of always-on agent design patterns.
Always-On Design Patterns Summary
Reliable always-on agents combine: daemon processes with signal handling for clean shutdown, watchdogs for automatic restart on crash, persistent WebSocket connections with exponential backoff reconnection, heartbeat checks to detect stuck processes, graceful shutdown to complete in-flight work, and state checkpointing to resume after restart. Use systemd or supervisord for production process management.
Frequently asked questions
Is the “Always-On Agent Design Patterns” lesson free?
Yes — the full text of “Always-On Agent Design Patterns” is free to read here on the web, and the AI Agents course includes 4 lessons in total. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the AI Agents course, upgrade to CoddyKit PRO.
What will I learn in “Always-On Agent Design Patterns”?
Background processes, daemon agents, and persistent connection management. You practise AI Agents with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.
Do I need any experience to start AI Agents?
No prior experience is required. AI Agents on CoddyKit is structured for beginners through advanced learners; this is — lesson 1 of 4, so you can start here or from the beginning and move at your own pace.
How long does the “Always-On Agent Design Patterns” lesson take?
Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.
Can I write and run code in this AI Agents lesson?
Yes. Every AI Agents lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.
All lessons in this course
- Always-On Agent Design Patterns
- Proactive Notification and Alert Systems
- Context Persistence Across Sessions
- Building a Daily Briefing Agent