Dev.to Security 🔐 Cybersecurity 👁 0 📖 3 min read

I got tired of self-hosting Wazuh and paying enterprise EDR prices, so I built my own Linux host agent

f you run a handful of Linux servers for a small team and nobody's full-time job is "security" your options for host defense kind of suck. On one end, you have Wazuh. It's free and powerful, which is great until you real

f you run a handful of Linux servers for a small team and nobody's full-time job is "security" your options for host defense kind of suck.
On one end, you have Wazuh. It's free and powerful, which is great until you realize you now have a second full-time job hosting, tuning, and babysitting an ELK/OpenSearch stack just to figure out why an alert fired. On the other end, you have CrowdStrike or Datadog security add-ons priced for enterprise compliance budgets, built for fleets of 5,000 endpoints, and usually handing you opaque risk scores instead of telling you what actually happened on the box.
I wanted something in the middle: a single lightweight binary I could drop onto a Linux server or Docker host in 60 seconds that watches processes, network connections, files, and SSH logins, explains what's going on in plain English, and takes bounded, reversible actions without bricking production at 3 AM.
So I built Vigilhold.

The core problem with "autonomous" security tools

Most developers don't trust automated remediation on production Linux boxes, and honestly, they shouldn't. If an agent hallucinates a threat and kills systemd or locks out your deploy user's IP, your security tool just caused an outage.
When designing Vigilhold (and its backend decision engine, Cortex), I set three hard rules:

  1. Strict autonomy boundaries. You configure what the agent is allowed to do on its own, host by host. By default, it will automatically ban obvious attacking IPs (like SSH brute-forcers), while anything disruptive killing a process, quarantining a file, isolating a host stops and asks for human approval first (you can approve or reject straight from Slack or the dashboard). If you want full Autopilot on non-critical boxes, you can turn it on.
  2. Everything has an undo button. Bans, file quarantines, and lockdowns are designed to be reversible, and every single step is written to a tamper-evident audit log. When you override or correct a decision, future decisions on that server actually take your feedback into account.
  3. Offline survival. If the backend goes unreachable, the local agent doesn't go blind it keeps blocking SSH brute-force attempts and already-flagged bad actors locally within your configured limits.

Don't trust my marketing watch a live box get attacked

Security landing pages are full of fake dashboard screenshots and "AI-powered threat theater." I hate that stuff.
Instead, I spun up a disposable Ubuntu VPS on the public internet, installed the Vigilhold agent on it, and made the feed completely public at vigilhold.com/lab.
There are no customer workloads on it just a bare public IP taking real hits from bots and scanners every few minutes. You can watch the raw SSH attempts, process signals, and automated blocks roll in live (delayed by 30 minutes to the nearest quarter-hour so attackers can't use the feed to time the box).
If you want to test it yourself without creating an account first, there's also an interactive sandbox at vigilhold.com/demo where you can trigger simulated threats and see how the plain-language breakdown and IP banning behave.

Under the hood & how it runs

Installing the agent is an outbound-only one-liner (no firewall ports to open, and a single command cleanly uninstalls it):

curl -fsSL https://vigilhold.com/install | sudo bash -s -- --token <your-token>

Once it's running:
• Signals watched: Process execution, network sockets, file integrity changes, SSH auth logs, and Docker container activity.
• No hosting overhead: You don't manage collectors, indexers, or rule pipelines.
• Clear shipping status: Because I'm building this independently (no VC bloat, just shipping code out of Utah at Al's-Ops), I keep a public Feature Status page labeling every single capability as LIVE, BETA, or ROADMAP so nobody has to guess what actually works today.

Try it out (or roast my architecture)

If you manage 1 to 25 Linux boxes and want host security without hiring a SOC analyst or maintaining an Elastic cluster, you can spin up a 14-day free trial (no credit card required) at vigilhold.com. After that, it's $39/month for up to 5 servers.
If you're a Linux sysadmin or security engineer, I'd genuinely love your feedback:
• What's the one automated action you would never let a host agent take on your servers without human sign-off?
• If you check out the live public lab or demo, what signal or context is missing that you'd want to see during an incident?
Drop a comment below - I'll be hanging around all day to answer questions about how the agent and rule scoring work.

📰 Read the original article on Dev.to Security

Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.