Paranoid~RV

Two weeks in. 23 containers. $0 hardware. Built from e-waste in a 40ft fifth wheel.

Here's what's running and where it's headed.

WHAT'S LIVE:
Suricata + Zeek — dual IDS watching every packet
Wazuh — SIEM + forensics
CrowdSec — crowdsourced threat intel
AdGuard + Unbound — private DNS chain
Node-RED + MQTT — automation nervous system and sensor bus
Local LLMs on a salvaged GTX 1050 Ti
Custom AI memory system — persistent context across sessions
TTS engine — she talks back
Grafana + InfluxDB — dashboards and telemetry

WHAT'S COMING — THE RV BECOMES SENTIENT:
Every light, shade, lock, and climate system on the MQTT bus. ESPHome sensors on every surface — doors, windows, water tanks, chassis vibration, air quality. Zigbee mesh throughout. Frigate NVR with local AI object detection on cameras. Presence tracking via Bluetooth beacons — the AI knows which room you're in and adapts around you.

Power monitoring on every circuit. Weather station on the roof feeding solar predictions. Water quality sensors on the freshwater input. TPMS on every wheel. GPS geofencing so the rig starts warming up when the truck gets close.

All local. All self-hosted. All orchestrated by one AI brain that never sleeps.

BUT HERE'S THE PART THAT KEEPS ME UP AT NIGHT:

A weatherproof helipad and charging station on the roof. A coordinated drone fleet. Not one drone doing laps — multiple birds working together, redistributing tasks if one returns to charge.

Drone-to-drone mesh communication. They talk to each other. If comms drop, failsafe protocols bring them home.

Routine perimeter inspections of my own property on schedule. Spot checks triggered by IDS alerts. Aerial inspection when the network sees something unusual. Systematic grid mapping with thermal imaging. Site photogrammetry when parked somewhere new — build a local map before I even step outside.

Ground-deployed sensor packages — temp, air quality, gas detection. Instant remote monitoring posts without leaving the rig.

The IDS sees an anomaly. The AI evaluates. Drones launch under operator authority. Live feeds processed on-stack. Drones dock, charge, wait.

A 40ft fifth wheel with its own air force. Controlled by an AI that runs on a machine someone threw in the trash.

All on cellular. No fixed address. No cloud. Self-hosted everything on a machine someone threw away. Self-taught everything from scratch.

No reason. Just wanted to see if I could.

#homelab #selfhosted #infosec #drones #AI #IoT #automation #linux #RVlife

Mar 21 at 11:59pmWeb
Show threadHakuso2d ago

@ParanoidRV
Do you want Maximum Overdrive?

Because this is how we get Maximum Overdrive.

Show threadParanoid~RV2d ago

@Hakuso Ha! Funny you say that — I was researching mesh connectivity between the truck and trailer, went down a rabbit hole on the biggest truck you can legally drive without a CDL, and yeah... Maximum Overdrive came up in that conversation too.

Love the reference. But I'm thinking less Maximum Overdrive, more Chappie. Teach the machine. Give it rules. Give it purpose. It becomes something that protects instead of destroys.

That's the whole build.

Show threadHakuso2d ago

@ParanoidRV
Well, I mean that was a serial killer's ghost IIRC, so more of a less creepy Chucky than HAL.

Still had that clown face in my head with all vehicle automation, even though Ghost in the Shell was probably more likely.

Show threadMiroslav Stankic1d ago
@ParanoidRV what’s your LLM setup? Have and old 1060 I wanted to play with but GoogleFu tells me bad things.
Show threadParanoid~RV12h ago
@pax0707 Hey Miroslav, your 1060 is actually better than what's running here. I've got a salvaged GTX 1050 Ti with 4GB VRAM doing the heavy lifting... Stack is LiteLLM as the routing layer with three tiers: DeepSeek r1 8b for fast responses, r1 14b for anything that needs to think, and Claude Sonnet via API for the hard stuff. Open WebUI on top... For your 1060 with 6GB, the 8b runs comfortably. The 14b is tight but doable with num_gpu=20 and num_ctx=1024. Worth confirming it fits before committing. Whatever Google told you bad things about was probably the 70b models or old CUDA drivers, neither applies here... What's the use case?