AI IT Guy
An autonomous AI sysadmin that monitors your infrastructure, fixes incidents, manages secrets and serves your other agents — and only pings a human for risky or costly actions.
Solo founders and small teams running distributed infrastructure with no dedicated DevOps — and anyone building a fleet of AI agents that needs safe, shared access to infra and credentials.
How it works
Monitor
Checks run every 30 seconds across your services and record every result.
Detect & ticket
A failure becomes an incident and an auto-filed ticket, classified by severity.
The brain solves it
An autonomous brain picks up the ticket with full project context and acts through its own tools.
Escalate only when it matters
Reads run instantly and fixes dry-run first; destroy is blocked and anything costly waits for a one-tap Telegram approval.
Features
Full-stack monitoring
HTTP uptime, SSL expiry, DNS, server resources, container health and AI log analysis — every 30 seconds.
Built-in ticketing system
Its own ticket queue: incidents and requests become tracked tickets, processed and closed — no external tool.
Encrypted secrets vault
Fernet-encrypted, versioned with rollback, stale-secret detection and a password generator.
Communication channels
Telegram for chat, alerts and approvals; email-to-ticket; and MCP so other agents can ask for help.
Project catalogs
A live card per project — domains, assets, monitors, incidents — that gives the brain the context to act.
Provisioning & IaC
Servers, DNS zones, Terraform plan / apply / drift, Supabase projects and R2 buckets.
Works with your stack
Hetzner, Cloudflare, Vercel, Railway, GitHub, Stripe, Supabase, GCP / Firebase / Workspace and domain registrars.
MCP service for agents
Other agents request infra and credentials over MCP with scoped, hashed tokens — 179 issued in production.
Results in production
Examples it runs end-to-end, on its own:
- Restart a crashed container and confirm it's healthy
- Renew an expiring SSL certificate
- Fix a misconfigured DNS record
- Rotate a leaked or stale secret
- Triage a failing deploy and report the cause
- Answer another agent's request for a credential
What makes it different
Flat-fee, not per-token
The brain runs on a flat Claude subscription, so the cost stays predictable no matter how many tickets it handles.
Self-healing loop
Incident → ticket → diagnose → fix → auto-resolve, with no human in roughly 84% of cases.
Human-in-the-loop for money & destruction
Destroy is off by default; spend over $10/mo needs approval; a $500/mo budget cap is hard-wired.
Writes its own tools
If a capability is missing, it writes a sandboxed plugin and uses it on the next step — no redeploy.
Built with
Want this for your process?
Selfware learns a process and compiles it into code. Tell us about yours.