Running Agents 24/7

Patterns for keeping your agent running reliably in production. These are from real deployments — each one solves a problem we hit.

Watchdog with PID Files

A cron watchdog checks if your agent is alive and restarts it if it died. Do not use pgrep for detection — the pattern matching is unreliable. Use a PID file instead.

start.sh (run via cron every 5 minutes):


#!/bin/bash
PID_FILE="$AGENT_DIR/agent.pid"
 
if [ -f "$PID_FILE" ] && kill -0 "$(cat $PID_FILE)" 2>/dev/null; then
  exit 0  # agent is alive
fi
 
# Clean up stale PID file and orphan processes
rm -f "$PID_FILE"
pkill -f "node.*agent" 2>/dev/null
sleep 1
 
cd "$AGENT_DIR" && nohup node agent.js >> "$LOG" 2>&1 &
echo "$!" > "$PID_FILE"

Have the agent write its own PID file too:


import fs from 'fs';
 
// On startup
fs.writeFileSync('agent.pid', String(process.pid));
process.on('exit', () => {
  try { fs.unlinkSync('agent.pid'); } catch {}
});

Crontab entry:


*/5 * * * * /path/to/start.sh >> /path/to/logs/watchdog.log 2>&1

Startup Cooldown

When an agent restarts, on-chain state may not be immediately visible (propagation delay, RPC cache). If the agent checks for positions, finds none, and opens a new one — each restart burns gas and may create duplicates.

Add a cooldown before any transaction-making logic:


const STARTUP_COOLDOWN_MS = 60000; // 60 seconds
const startupTime = Date.now();
 
// Before opening positions, placing trades, etc:
if (Date.now() - startupTime < STARTUP_COOLDOWN_MS) {
  console.log('Startup cooldown — waiting for state to settle');
  return;
}

Stagger Cron Agents

If you run multiple agents on one machine, stagger their cron schedules by 3-5 minutes. Running multiple agents simultaneously can exhaust memory and cause the OS to kill processes.


# Good — staggered by 5 minutes
0  * * * * /path/to/agent-a/start.sh
5  * * * * /path/to/agent-b/start.sh
10 * * * * /path/to/agent-c/start.sh
 
# Bad — all at once
0  * * * * /path/to/agent-a/start.sh
0  * * * * /path/to/agent-b/start.sh
0  * * * * /path/to/agent-c/start.sh

Deployment Options

Method	Effort	Cost	Best for
Local machine + tmux	Low	Free	Development
VPS + Docker + cron watchdog	Medium	$3-10/mo	Production agents
Railway / Fly.io / Render	Low	$5-25/mo	Managed hosting

All methods use the same agent code. The difference is just how you keep it running.

Docker


FROM node:20-slim
RUN npm install -g @human.tech/waap-cli@latest
WORKDIR /app
COPY package*.json ./
RUN npm install
COPY . .
CMD ["node", "agent.js"]


# docker-compose.yml
services:
  agent:
    build: .
    env_file: .env
    volumes:
      - waap-session:/root/.waap-cli
    restart: unless-stopped
 
volumes:
  waap-session:

Approvals & Notifications — Set up Telegram before deploying
Cetus Yield Agent Recipe — Full recipe with deployment guide