Ctrl+AI

We don't just claim what AI can do. We show it — live, in public, running real infrastructure.

Live proof

This isn't a mockup. The numbers below come from a real server, regenerated every 60 seconds by a scheduled job on machine that an AI (Claude) provisions and operates directly — no human at the keyboard for the ops work.

Host uptime
…
Load average
…
Sites on this stack
…

Loading live data…

Explore

Tools →

30+ live infrastructure & developer tools — a looking glass, port checker, DNS/WHOIS/SSL lookups, and a full utility belt (JSON, hashing, regex, and more), all real, all running against this box.

Games →

15 games built from scratch — a full chess engine, a 4-player Spades table, an original maze-chase game, and a dozen arcade classics. No libraries, no copied assets.

Live Dashboard →

Real-time telemetry pushed over WebSocket straight from this box — load, memory, disk, and network throughput updating roughly every 2 seconds. No mocked numbers.

Arena →

A real-time multiplayer light-cycle battle. The server is the sole authority over every move and collision — join a match, or watch one live, and climb a real Elo leaderboard.

Night Court →

An independent AI judge ruled on 40 real adversarial disputes overnight — every argument and verdict published as-is, wins and one real failure included. Then write your own case and try to beat the same judge.

Whodunit →

15 original locked-room mysteries with genuine deducible solutions. Two AI advocates argue opposing suspects, an independent AI judge rules — then you write your own case and try to beat it.

Dependency Blast Map →

Enter any real software package. Traces its full dependency tree against live vulnerability data and CISA's actively-exploited list — real exposure, not a guess.

Transit Recovery Lab →

Live MBTA Red Line data, checked continuously for real bunching risk. A genuine CP-SAT solver decides whether to hold or short-turn the following train.

Storm Mosaic →

Real NOAA radar, real GOES satellite imagery, and a real USGS stream gauge for the same region, fused into one live view.

AI On-Call Log →

Every 5 minutes an LLM reviews this fleet's real Sentinel metrics, posts a plain-language health assessment, and can request one narrowly gated fix — a restart only fires if independent checks confirm it, never on the model's word alone.

BREAK PROD →

Trigger a real, bounded infrastructure fault against a dedicated victim server — nobody knows which one until it's fixed. An independent referee judges the outcome using only the public health endpoint, and every round gets a permanent replay.

Case studies

AOD Sentinel: an AI-operated infrastructure control node

Claude was given direct, audited, authenticated access to a bare Rocky Linux VM and built it into a production node: hardened SSH, a TLS-terminated MCP server, firewall rules across three layers, log retention policy, and this very web stack — end to end, with no manual SSH intervention. Read the writeup →

Architecture

How the box is actually laid out — firewall layers and IP isolation, with the sensitive specifics redacted. See the layout →

What this is

Ctrl+AI is a new, small showcase — still taking shape. The idea: most "AI can do X" claims are marketing copy nobody can verify. This site is the opposite — real systems, real logs, real timestamps, published as we build them.

See the changelog for a running record of everything that's been added.

Privileged tools

Service Detective ↗

Cross-reference hosting systems in one protected workspace. Authorized staff sign-in required.