Edge AI monitoring for small buildings and equipment. Predict failures before they happen. Off-the-shelf sensors, local inference, no cloud lock-in.
Book a ConsultationCustom AI monitoring systems for buildings and equipment that the big vendors ignore.
Smart plugs, environmental sensors, and energy monitoring across your entire facility. One dashboard. AI-powered anomaly detection. Know when something is wrong before it breaks.
Sensors on your HVAC, pumps, and electrical systems feed a local AI that learns normal patterns and alerts you when something drifts. Prevent the $8,000 emergency call.
Done-for-you Home Assistant setup. Lighting, climate, security, energy monitoring. Professional configuration, ongoing management, no subscription to big tech platforms.
Unique monitoring needs? Wine cellars, server rooms, grow operations, fleet vehicles. We design and deploy custom sensor networks with AI analysis.
Real systems running in production right now.
A $25 smart plug, a local LLM, and three USGS stream gauges. Detects stuck floats, runs automated recovery, correlates with weather and river levels. Running 24/7, prevented a basement flood.
Read the full case study →Home Assistant managing smart plugs, thermostats, printers, mesh WiFi, and speakers across 8 zones in a community building. Energy monitoring, automated scheduling, single dashboard.
Learn more →Human pose estimation using WiFi signal analysis instead of cameras. No privacy concerns. Works through walls. Rust implementation with neural inference pipeline. Applications in eldercare and security.
Learn more →No proprietary hardware. Smart plugs, environmental sensors, and existing equipment data. Total hardware cost: usually under $500.
Your data stays in your building. We run LLMs on local hardware (Mac, Linux, Raspberry Pi). No cloud subscriptions. No internet dependency. Works during outages.
Safety-critical systems need human oversight. Our AI layer detects anomalies, correlates patterns, and sends actionable alerts. Automated recovery protocols have manual overrides built in.
Home Assistant, MQTT, REST APIs, SQLite. No vendor lock-in. You own your data and your system. We build it, you keep it.
What this actually looks like week to week, written up as it happens.
AGIBOT proposed Worken, a way to count the useful work a robot does instead of its powered-on time. I ran it against my sump pump: 669 motor hours since February, 4.3 started by the float, and 24 days with no evidence any water was moved.
October 30, 2026Symbotic gave Walmart the right to escrowed technical material and engineers on site. I ran the same test on my own sump pump monitor: the code running it was last committed in June, the edit that hid 8,386 dropped alerts exists on one disk, and the fix for an 18-hour outage disappears on the next reboot.
October 29, 2026Every machine health score I read about this week starts from the motor's power draw. For three days last week my sump pump drew 45 watts less than usual, and four layers of monitoring called it healthy. They were right about the pump. The utility had dropped three volts. Across 180 days the pump's watts track the line voltage with r = 0.96, and once you subtract it the motor has not moved in eight months.
October 28, 2026At NexusCon, a vendor asked whether the useful little app a building engineer stands up is still running a year later. Mine is: 10,093 log lines in eleven days, ten daily digests, every one OK. It also pulsed 119 times a day through ten rainless days exactly as it did through a storm, sent its only urgent alert about the utility's voltage six times in 55 minutes, and has two one-line fixes filed eleven days ago that nobody applied.
October 27, 2026A practical vision for AI agents in smaller cities: business capacity, community projects, responsible pilots, funding routes, and a longer view of engineering and discovery.
October 19, 2026The Linux kernel measured its AI code reviewer against 1,000 known bugs before leaning on it. I had never measured mine. Thirty notes, seven days: the watcher reads thirty lines of log, reads them accurately, and labels the result a seven-hour review. Twice it was handed a count of five failed reads and wrote that the plug was stable.
October 16, 2026Start with a $25 sensor. End with a building that thinks.
Book a Consultation