Your AI agent says it's done.
Prove it.
I audit and monitor AI agents running on your servers. Read-only scan in minutes. Full manual audit in 48 hours. I have the authority to stop the agent when something goes wrong.
What I Check
Nine security controls that every AI agent host needs. The automated scan checks all nine in seconds. The manual audit goes deeper on each one.
⚿ Secrets Exposure
.env files readable by other users or served from your web root. API keys the agent can reach but shouldn't.
⊘ Agent Privileges
AI agents running as root with full system access. No sandboxing, no user isolation, no resource limits.
⚡ SSH & Remote Access
Root login enabled, password auth on, no key rotation. The door the agent — and attackers — walk through.
▦ Firewall & Brute-Force Guard
No firewall active. No fail2ban or CrowdSec. Every port open to the internet by default.
▤ Command Logging
No record of what the agent ran. No auditd rules, no command wrapper. You'll never know what it did.
⏻ Kill Switch
No way to stop the agent that doesn't rely on the agent itself. If it goes rogue, you're watching.
↺ Tested Backups
Backup jobs that exist but were never restored. Backups that aren't off-server. Backups the agent could delete.
◎ Network Exposure
Ollama, Qdrant, n8n dashboards bound to 0.0.0.0. Internal services accessible from the public internet.
▣ Disk Gate & Runaway Protection
No disk usage limits. No circuit breaker when the agent fills the disk. No memory or CPU quotas.
How It Works
From first scan to full audit in four steps. You stay in control the entire time.
Free Scan
Run the read-only scan script on your server. Takes under 30 seconds. Nothing is installed or changed. You get an instant severity summary.
Review Results
I review the scan output, flag what matters, and send you a clear summary with the risks ranked by severity.
Full Audit
I go deeper on every finding. Manual checks, configuration review, and a written report with exact fix commands and acceptance tests.
Monitor & Protect
Ongoing monitoring retainer: command logging, kill switch, disk gate, watchdog alerts. I can stop the agent when it goes wrong.
Pricing
Transparent pricing. No surprise charges. Every engagement starts with the free scan.
Rapid Audit
Scan review + priority call
One-time · 24-hour delivery
- Automated scan + manual review
- 1-hour prioritization call
- Written fix list with commands
- Email support for 14 days
Full Audit
Complete security assessment
One-time · 48-hour delivery
- Everything in Rapid Audit
- Deep configuration review
- Written report with severity ratings
- Fix commands + acceptance tests
- 30-day re-test after fixes applied
- Kill switch & command logging setup guide
Monitoring
Ongoing protection retainer
Minimum 3 months · billed monthly
- Everything in Full Audit
- Command logging installed & monitored
- Kill switch with alert routing
- Disk gate & runaway protection
- Monthly security report
- Authority to stop the agent on threats
Questions
Straight answers. If yours isn't here, ask me.
What access do you need?
For the free scan: none — you run the script yourself and send me the output. For the paid audit: read-only SSH access for a temporary user, or the scan output plus answers to follow-up questions. I never need root credentials. You remove the access when the audit is done.
Do you touch any data or change settings?
No. The audit is read-only. I read configuration files, process lists, listening ports, file permissions, and log metadata. I don't install software, read application data, or run exploits. The report tells you what to fix and gives you the exact commands.
What AI agents do you audit?
Any agent running on your server: OpenClaw, Hermes, AutoGPT, CrewAI, LangGraph, n8n, Ollama, aider, open-interpreter, Claude Code, Codex, custom agents — anything with a process ID and shell access. If it can run commands on your server, I can audit it.
What does "kill switch authority" mean?
Monitoring subscribers authorize me to stop the agent (via kill switch, service stop, or key revocation) when I reasonably believe it's deleting data, exposing secrets, running out of control, or has been compromised. I notify you immediately after acting and never delete anything myself.
What frameworks do you align with?
The audit maps findings to OWASP's Top 10 for Agentic AI (ASI01–ASI10), the EU AI Act's requirements for high-risk AI systems (audit trails, logging, human oversight), and NIST AI RMF. The report notes which framework each finding relates to.
How is this different from a penetration test?
A pentest tries to break in from outside. This audit checks the controls around the AI agent you've already invited inside your server. The risk profile is different: the agent has legitimate access, so the question is whether that access is properly scoped, logged, and revocable.
Can I see a sample report?
Yes. Request one and I'll send a redacted sample from a previous audit. No client data, just the structure, severity ratings, fix commands, and acceptance tests.
Start with the free scan
Run the script. See the results. Decide if you want me to go deeper.
Get Started
Tell me about your server and the agents running on it. I'll send the scan script and get back within 4 hours.