I build security teams. Lately, some of them aren’t human.

Most small companies will never have a SOC. They can’t staff one around the clock and can’t pay for the tooling. Mini AI SOC is my attempt to fit the whole thing, SIEM, threat intelligence, SOAR, network IDS, a TLS-inspecting proxy and a honeypot, into one stack that starts with a single command.

An AI analyst works every alert with tool calls before a person looks at it. It also writes detection rules from a plain-English description, tunes the noisy ones, and points out which attacks your rule set doesn’t cover yet. The same platform runs as a managed service for small and mid-size businesses, with human analysts available on top.

miniaisoc.com

In the box

• 170+ detection rules, mapped to MITRE ATT&CK
• 34 log formats, normalised to Elastic Common Schema
• Microsoft 365, Entra ID and Defender, across tenants
• 35 threat-intel feed presets, TAXII and STIX
• Suricata IDS, Squid with ClamAV, Cowrie honeypot
• Endpoint agents for Windows and macOS
• SOAR playbooks that fire on new alerts

Every employee has a name, a job description and a shift. A creative director reads what’s trending and pitches one game a night. A game creator builds it end to end, from intro screen to game over. Others finish, fix and tidy up older games. Nothing counts until an independent QA agent has loaded the game and played it.

In the morning the orchestrator rebuilds the portal and writes the devlog, the marketer posts what shipped, and I review the diffs and commit. Studio HQ, the management panel, shows the staff at their desks in a four-storey 3D office, which also works in a VR headset.

It’s the same thing I do at work: clear roles, a clear quality bar, and someone checking the work. Only the team is different.

vibegamesstudio.comMeet the staff

The night shift

01:00	Game ImproverPushes three older games closer to finished.
02:00	ProducerMoves the flagship game one milestone forward.
02:30	Creative DirectorPitches one new game from the trend radar.
03:00	Game CreatorBuilds that game, start to finish.
03:40	Game MedicReproduces and fixes real bugs in one game.
06:30	OrchestratorRebuilds the portal, writes the devlog.
07:00	MarketerPosts to Discord, Bluesky and Mastodon.
Morning	MeReads the diffs. Commits.

This is phase one of a robot that will eventually drive around the house on wheels. The head comes first, and runs the same on my PC as it will on a Raspberry Pi 5.

It wakes on its own custom-trained wake word, then streams the reply sentence by sentence, so it starts talking before it has finished thinking. Each reply is a single JSON object that drives the voice, the face and where it looks, all at once:

{ "emotion": "curious",
  "speech": "Wat doe jij daar?",
  "look_at": "left" }

It keeps its own memory as a folder of Markdown notes that it reads and writes with tools. And it runs the house through its own MCP server: Tuya lamps, a Roborock vacuum and a Honeywell thermostat, all by voice.

“Robocop, doe de woonkamer maar paars.”

Private repo for now. It knows the keys to my lamps.

What it does

• Wakes on its own name, with a model I trained locally
• Hears and speaks Dutch; the Piper voice runs offline
• One conversation per day, each starting from a summary of yesterday
• Long-term memory it organises itself
• Lights, vacuum and thermostat over MCP
• A face for any screen that dims at night and dozes off when idle