Skip to content
Back to writing

~/writing field note

  • openclaw
  • vps
  • self-hosting
  • hetzner
  • ai-agents

Run OpenClaw on a $5 VPS: Full Setup, Fixes, and Real-World Results

Mustafa Ramx pixel avatarBy Mustafa Ramx18 Feb 20266 min readUpdated 21 Feb 2026
Cover image for Run OpenClaw on a $5 VPS: Full Setup, Fixes, and Real-World Results
On this page10 sections
  1. How to run OpenClaw on a cheap VPS (step-by-step)
  2. What went wrong first with OpenClaw (and why people think you need expensive hardware)
  3. The fixes that made it stable
  4. GUI optional, not always-on
  5. OpenClaw browser automation reality (important)
  6. “AI modules” I used in OpenClaw
  7. Useful OpenClaw prompts that worked for me
  8. Step-by-step blueprint (copy this if you’re building now)
  9. Honest pros vs cons: $5 VPS vs Mac Mini
  10. Final take

I ran OpenClaw on a $5 VPS (while everyone said “you need a $500 Mac Mini”).

Quick answer first: yes, it works.

Everyone online says:

You need a Mac Mini to run OpenClaw properly!

As a server administrator and DevOps expert, I tested the opposite.

I ran OpenClaw on a $5/month VPS (Ubuntu) and got it working for:

  • Daily stock market reports
  • Scheduled motivational messages
  • Telegram automation
  • Browser control / testing
  • Subagent workflows

Not perfect at first. But after a few real fixes, it became stable enough for serious usage.

If you want to copy this setup, this article will save you hours. 😉

How to run OpenClaw on a cheap VPS (step-by-step)

Here’s a short walkthrough of the actual build before we get into the details:

My VPS setup (real stack)

  • Provider: Hetzner Cloud
  • Server size: cx23 small (2 vCPU / ~4 GB RAM class, 40 GB SSD) — upgradeable
  • OS: Ubuntu 24.04
  • OpenClaw: latest stable (updated during setup)
  • Main channel: Telegram
  • Extras used: cron jobs, subagents, chrome relay, managed browser mode
  • Docker to run n8n for other testing
  • SSH default port changed for security
  • Hetzner firewall on with only the SSH port opened for more security
  • SSH password off, connected by the private key

If you want to support this blog, use my Hetzner referral link: https://hetzner.cloud/?ref=g1aTOgwu1xST

What I wanted OpenClaw to do (real use case)

I didn’t want a toy demo. I wanted production-like daily value:

  1. Morning stock market & crypto analytics & brief
  2. Post-open US stock market update
  3. Daily “Morning Soul” motivational message
  4. Reliable Telegram delivery
  5. Automated social media marketing
  6. Full-time job searching and applying

This is where most guides stop. I’m sharing what happens after install.

What went wrong first with OpenClaw (and why people think you need expensive hardware)

At one point the VPS became almost unusable:

  • CPU looked “100%” because of n8n and Docker
  • Box felt frozen

Actual root cause:

It wasn’t pure CPU. It was memory pressure + no swap → kernel thrashing → system CPU spike.

That’s the hidden killer on cheap VPS plans.

The fixes that made it stable

1. Add swap (most important)

I added a 4 GB swapfile and persisted it in /etc/fstab.

This immediately reduced crash risk and helped absorb spikes.

2. Tune memory behavior

I set:

vm.swappiness        = 10
vm.vfs_cache_pressure = 50

This improved stability during heavy mixed workloads.

3. Disable services I didn’t need

  • Docker + containerd disabled (for now) — stopped the need for n8n; OpenClaw did all the needs ;)
  • CUPS service disabled (not needed on this host)
  • Kept only required services alive

4. Improve data-fetch cost and reliability

For recurring reports:

  • web_fetch first on known URLs
  • web_search Brave API only for discovery / fallback
  • Fixed source-map file for canonical links to get the stock market reports I need

This lowered API cost and reduced noisy output.

GUI optional, not always-on

Installed a lightweight GUI (Openbox + TigerVNC + noVNC), but:

  • Disabled auto-start
  • Run only when needed — only to automate logged-in tasks like publishing on x.com, Reddit, or here, like what you’re reading right now 😎

So I get GUI power when needed, without paying the RAM cost 24/7, and it’s accessible by VNC.

OpenClaw browser automation reality (important)

There are two modes in OpenClaw:

A) Managed browser, headless (stable) — great for scraping / research flows, deterministic runs.

B) Chrome relay GUI (needed for real account login) — needed when you must use your own logged-in browser session (X / Reddit, Facebook, etc.).

I found a practical hybrid:

  • Login lane: Chrome relay
  • Automation lane: managed browser

That gave me both reliability and real-world login capability.

“AI modules” I used in OpenClaw

If you’re new, this is the core stack I actually used:

  • gpt-5.3-codex (OAuth login, monthly subscription, no extra credits)
  • Gateway (main service)
  • Cron jobs (scheduled tasks)
  • Subagents (isolated long / complex runs)
  • Browser tool (profile="chrome" and profile="openclaw")
  • Memory files (persistent context)
  • Telegram messaging / reactions / inline buttons
  • Session orchestration tools

This is enough to build a real assistant workflow, not just chat.

Useful OpenClaw prompts that worked for me

1) Host-diagnosis prompt

Check this VPS for performance bottlenecks in the last 2 hours:
- CPU, RAM, swap, load, disk IO
- kernel logs for memory pressure / OOM
- top processes by CPU and RAM

Then give:
1) root cause
2) immediate fixes
3) safe commands (with explanation)

2) Cost-optimization prompt

Refactor my recurring report workflow to:
- use web_fetch first for known URLs
- use web_search only for discovery / fallback
- add a source-map file
- enforce max search calls per run

Then update prompt files and summarize changes.

3) Browser-mode-strategy prompt

Design a dual-browser operation:
- managed browser for stable automation
- chrome relay for human login / 2FA flows

Include exact operating rules to reduce relay drops.

Step-by-step blueprint (copy this if you’re building now)

  1. Deploy Ubuntu (no GUI) VPS on Hetzner.
  2. Install OpenClaw + verify gateway.
  3. Connect Telegram.
  4. Create first cron job (simple text).
  5. Add swap + memory tuning.
  6. Disable unused services (Docker / CUPS if not needed).
  7. Build one real recurring workflow.
  8. Add source-map + fetch-first policy.
  9. Add subagent fallback for heavy tasks.
  10. Add optional lightweight GUI (manual-start only).

Honest pros vs cons: $5 VPS vs Mac Mini

VPS wins:

  • Super cheap
  • Remote-first
  • Automations run 24/7
  • Easy to rebuild / scale

Mac Mini wins:

  • Smoother browser / UI workflows
  • More local power headroom
  • Fewer relay edge-cases

My conclusion:

If you know DevOps basics, a $5 VPS is absolutely usable — and upgradeable as you need. If you want “no tuning, max smoothness,” a Mac Mini is easier.

Final take

This wasn’t a benchmark flex. This was a practical build to run real daily workflows with AI.

And it worked.

If you’re technical and budget-aware, you don’t need to wait for expensive hardware. Start with a cheap VPS, fix the bottlenecks properly, and ship.

Found this useful or inspiring? You can support my work — buy me a coffee. Every little bit helps keep the magic alive, and my brain awake 😅

// share this note

Share on XLinkedIn