Field notes
Working notes from the engineers who build it: running AI on hardware you own, where its limits are, and what we learned when something broke.
5 notesRSS
We rebuilt a 118 page site in a week. Here is what actually took the time
114 pages live with zero broken links and about forty plugins gone. The fast parts took a day. The slow parts are the ones worth reading about.
WebsitesCostWhat breaks in a live stream, and which of it you can engineer around
Cameras, cables, encoders, uplinks, power and the operator. Which failures a mixer can survive, which ones it cannot, and where the real risk sits.
Live videoReliabilityWhat an AI phone line costs, and when owning the hardware would make sense
The monthly plans worked through at three call volumes against an answering service, and the volume where a private voice agent is worth a quote.
Voice agentCostPhoneCan an AI agent use software that has no API?
Yes, by working through the screen. How observe, act and verify actually run, what the loop measures at, and which tasks are bad candidates for it.
AI staffAutomationReliabilityWhat it actually costs to run a private language model in a small office
The honest arithmetic: the machine, the setup, the power, the seats you would have paid for instead, and the number of regular users where owning starts to win.
CostPrivate AIHardware
Tell us what is taking your team's time.
Thirty minutes with an engineer, free. You leave with a written page on what to automate first, and a written quote if you want us to build it.