A virtual assistant costs $1,000 to $4,000 per month. They work maybe 8 hours a day, need holidays, and miss things. An AI assistant running on OpenClaw costs you roughly $5/month in electricity and works every single second of every single day. Here is exactly how to set one up.
What You Will Learn
What "Always On" Actually Means
Most AI tools are reactive. You open ChatGPT, type something, get a response, close it. That is not an assistant. That is a search engine with personality.
An always-on AI assistant is different. It runs continuously in the background. It checks your email. It monitors your calendar. It publishes content on schedule. It responds to messages on Telegram, Discord, or WhatsApp the moment they arrive. Not when you remember to open an app.
OpenClaw is built for exactly this. It installs on hardware you own (or a VPS you rent), connects to your messaging platforms, and stays running 24/7 as a background service. No browser tab. No app to keep open. It just works.
Key difference: ChatGPT, Gemini, and other chat interfaces are session-based. You close the tab, they stop. OpenClaw runs as a daemon process. You could throw your laptop in the ocean and it would still be working on your Mac Mini at home.
Hardware: Mac Mini vs VPS vs Old Laptop
You need something that stays on. Three realistic options:
| Option | Upfront Cost | Monthly Cost | Best For |
|---|---|---|---|
| Mac Mini M4 | $499 | ~$3-5 (electricity) | Serious users who want local control |
| VPS (Hetzner, DigitalOcean) | $0 | $5-20 | People who want zero hardware |
| Old laptop | $0 | $8-15 (electricity) | Testing before committing |
The Mac Mini M4 is the winner for most people. According to real user measurements on Reddit, it draws about 4 watts at idle and 10-15 watts under load. That is less than a single LED light bulb. Running it 24/7 for a year costs less than $60 in electricity in most countries.
A VPS works great too. Hetzner offers ARM servers from around $5/month. DigitalOcean starts at $6/month. Both are enough for OpenClaw.
An old laptop works in a pinch but is unreliable for true 24/7 operation. Batteries degrade, fans collect dust, and they were not designed to run non-stop.
My setup: I run OpenClaw on a Mac Mini M4 sitting on my desk. It is silent, always on, and I forget it is there until I see what it accomplished overnight.
Heartbeat and Cron: The Two Engines of Always-On
OpenClaw has two systems that make it proactive instead of reactive:
Heartbeat is a periodic check-in. Every 30 minutes (configurable), OpenClaw wakes up, reads a checklist you define, and decides if anything needs attention. New emails? Calendar reminder? Social notification? It handles it or alerts you.
Cron jobs run at exact times. "Post to X at 9 AM." "Send a morning briefing at 7 AM." "Publish a blog article at 9 PM." These fire on schedule, every time, no matter what.
Heartbeat vs Cron: Heartbeat is like checking your phone periodically. Cron is like setting an alarm. Use heartbeat for monitoring (email, notifications, weather). Use cron for scheduled actions (posting, reporting, reminders). Full cron setup guide here.
The combination is what makes it feel alive. Your agent checks for urgent things between scheduled tasks. Nothing falls through the cracks.
What a 24/7 AI Assistant Actually Does While You Sleep
This is not theoretical. Here is what my OpenClaw agent does on a typical night while I am asleep:
- 11:00 PM: Publishes scheduled social media posts to X
- 11:30 PM: Heartbeat check. Scans email for urgent messages. Finds nothing. Goes quiet.
- 12:00 AM: Runs overnight research on podcast guest prospects
- 6:00 AM: Compiles a morning briefing with calendar, email summary, and trending topics
- 7:00 AM: Sends the briefing to my Telegram. I wake up, read it, and know exactly what my day looks like.
That is 8 hours of work I never touched. No VA timezone issues. No miscommunication. No "sorry I missed that."
The overnight test: The best way to know if your AI assistant setup is working: go to sleep. Wake up. Check what it did. If the answer is "nothing," you have not configured it properly. If you wake up to a briefing, published posts, and organized inbox summaries, you are there.
Real Cost Breakdown: AI Agent vs Human VA
According to Wing Assistant and ZipRecruiter, a US-based virtual assistant costs $4,000 to $9,600 per month in 2026. Offshore VAs from the Philippines or Latin America run $1,000 to $3,000 per month. And they work 8 hours a day, 5 days a week.
Here is what a 24/7 OpenClaw setup actually costs:
| Expense | Monthly Cost |
|---|---|
| Mac Mini M4 (amortized over 3 years) | ~$14 |
| Electricity (4-15W, 24/7) | ~$3-5 |
| Claude API (Haiku for routine tasks) | $10-30 |
| Claude API (Sonnet/Opus for complex tasks) | $20-50 |
| Total | $47-99/month |
That is $47 to $99 per month for something that works 24 hours a day, 7 days a week, 365 days a year. Compare that to $1,000+ for a human who works a fraction of those hours.
The API costs depend on usage. Claude Haiku 4.5 costs $1 per million input tokens. Good enough for routine checks, heartbeats, and simple tasks. Claude Sonnet 4.5 at $3 per million input tokens handles complex writing, research, and decision-making. You mix and match based on what each task needs.
Be honest about the tradeoff: An AI agent is not a human. It cannot hop on a Zoom call for you, make judgment calls about sensitive client situations, or do physical tasks. It replaces the 80% of VA work that is repetitive: scheduling, research, drafting, monitoring, posting. The other 20% still needs a human. For most solo founders, that 80% is what burns their time.
How to Set It Up in Under 30 Minutes
The full setup from zero to always-on agent takes about 20-30 minutes:
- Install OpenClaw on your Mac Mini or VPS. One command: visit installopenclawnow.com for the install script.
- Add your API key. You need an Anthropic API key for Claude. We have a full guide for that. (Coming soon.)
- Connect a messaging platform. Telegram is the easiest to start with. Create a bot via BotFather, add the token to your config, done.
- Configure heartbeat. Set a check interval (30 minutes is a good start). Define what it should check: email, calendar, notifications.
- Add cron jobs. Start with a morning briefing at your wake-up time. Add more as you go.
- Write your SOUL.md and AGENTS.md. Tell the agent who you are, what you care about, and how it should behave. This is what makes it yours instead of generic.
That is it. From this point, it is running. You can message it on Telegram any time. It will respond instantly, check things proactively, and execute scheduled tasks on time.
Start small. Do not try to automate everything on day one. Start with a morning briefing and one social media post per day. Add more tasks as you trust it. Within a week you will wonder how you ever functioned without it.
Pro Tips for Running 24/7 Without Burning Money
Use cheaper models for routine tasks. Heartbeat checks do not need Opus. Set Haiku or Sonnet as the default model and only use Opus for complex work like writing or research.
Keep your HEARTBEAT.md file small. Every heartbeat check processes this file. If it is 500 lines, you are burning tokens every 30 minutes for no reason. Keep it under 20 lines.
Use cron for isolated tasks. Cron jobs run in their own session. They do not pollute your main conversation history. This keeps context clean and costs down.
Set up the memory system properly. MEMORY.md and daily memory files let your agent maintain context across sessions. Without them, it wakes up with amnesia every time. With them, it knows your preferences, your projects, your schedule.
Monitor your API spend. Anthropic and OpenAI dashboards show daily token usage. Check weekly. If something is burning through tokens, you probably have a chatty heartbeat or a cron job that is running too frequently.
Enable prevent-sleep on your Mac Mini. Go to System Settings, Energy, and make sure "Prevent automatic sleeping" is on. One setting. Do not skip it.
OpenClaw Lab is the #1 community for founders building AI agent systems. I share the exact playbooks, skill files, and workflows inside. Weekly lives, expert AMAs, and 265+ members building real systems.
Join OpenClaw Lab →