Something shifted in the last couple of years. People who would never call themselves self-hosters — who have never opened a terminal and have no interest in starting — are asking a version of the same question: can I own this thing outright?
You can watch it happen in any home-improvement forum. Someone bought four cameras, the app was fine for a year, and then the free clip history shrank, or the good detection moved behind a subscription, or a firmware update changed the app, or the account got locked and took the cameras with it. The complaint is rarely about money. It’s that a device sitting on their own wall stopped answering to them.
“Off-grid” is the word that keeps getting used for the fix. It’s a bad word for it, because it sounds like one binary switch — cloud or cabin — and that framing makes people either overbuild or give up. In practice, a camera setup has four independent grids, and you can step off them one at a time, in any order, and stop whenever the returns flatten out.
The four grids
1. Storage. Where the video actually lives. A vendor cloud, or a disk in your house.
2. Detection. Where the “is that a person” decision happens. Someone else’s GPU cluster, or a chip on your desk.
3. Access. How your phone reaches the video. Through a vendor’s servers by default, or through your own network.
4. Power and connectivity. Whether the thing survives an outage at all.
These are genuinely separate. Plenty of otherwise-cloud setups record locally to an SD card. Plenty of local NVRs still phone home for detection. Mixing is normal, and knowing which grid you’re actually on is most of the work.
The ladder, in value-per-effort order
Step 1: Get storage local. This is the big one and it’s the cheapest. Recording to a disk you own means retention is a hardware decision instead of a pricing-tier decision. A 4TB drive holds weeks of continuous 1080p from a handful of cameras, and it does not get shorter because someone reworked a plan page. Cost: you now own a disk that will eventually die, so you plan for that.
Step 2: Get detection local. Cloud detection means every clip your cameras consider interesting is uploaded, decoded and analyzed somewhere else. Running it locally — Frigate with a Coral TPU or a modest GPU is the common path — means the footage never leaves the building unless you send it. Cost: one evening of config, and you own the tuning. Thresholds, zones, and masks are now your job. (We wrote up how to tune those without guessing.)
Step 3: Get viewing local. On your own network, video should go camera → your server → your screen, with no third leg out to the internet and back. It’s faster, and it’s one fewer party in the path. Cost: basically nothing, if the software supports it.
Step 4: Get remote access private. This is the step people skip or do badly. The two bad answers are “port-forward 5000 to the internet” and “install the vendor cloud again.” The good answer is a private overlay network — Tailscale, WireGuard, or a VPN into your router. Your phone joins your network; nothing about your server is exposed to the public internet. Cost: an hour, once, and it’s the highest-leverage hour in this list.
Step 5: Get power independent. A small UPS on the server, the switch, and the router turns a 20-minute outage into a non-event and gives the machine time to shut down cleanly during a long one. This is the step most people never take and then regret exactly once, when they go looking for footage from a storm and find the recorder rebooted through the whole thing.
Most homes get almost everything they wanted from steps 1, 2 and 4. Steps 3 and 5 are polish.
What you don’t get back
An honest ladder has a bottom rung labeled this is what it costs you, so:
You are now the support team. When it breaks at 11pm, nobody is on call. That’s the actual trade — not money, not difficulty, but ownership of the failure.
Notifications still need the internet. This one surprises people. You can move storage, detection and viewing entirely in-house, but a push notification to a phone that is not on your network still has to travel through Apple’s or Google’s push service. Self-hosting removes the vendor’s servers from the path; it does not remove physics. Any system that tells you something happened while you’re at work depends on your internet connection, your phone’s connection, and a push service in between — all three of which can be slow or drop a message. Treat push as best-effort no matter who built it. If a message absolutely has to arrive, no consumer camera system of any architecture is the right tool.
Off-site copies are on you. A disk in the house is not a backup against the house. If that matters for your situation, that’s a deliberate second copy somewhere else, and it’s your job to set it up.
Hardware is a purchase, not a subscription — but it’s still a purchase. A mini-PC, a detection accelerator and a drive is a few hundred dollars up front. It usually pays back against subscription fees within a couple of years, and the math is worth doing honestly rather than assuming.
A ten-minute test worth running
If you already have a setup and you’re not sure which grids you’re on, unplug the WAN cable from your router — not the router, just its uplink — and leave it out for ten minutes. Then check:
- Are the cameras still recording?
- Did detection keep running and keep writing events?
- Can you still pull up live video from a device on your own network?
- When the internet comes back, is there a gap in the timeline?
Every “no” is a grid you thought you were off and aren’t. That’s a genuinely useful thing to know before the day you need the footage, and it costs nothing but ten minutes.
Where we sit in this
We build ViewPane, a viewer for Frigate — the phone app for a server you already run. We’re a small part of step 3 and step 4, and deliberately not part of steps 1 and 2, because those should never depend on us. If ViewPane disappeared tomorrow, your recordings, your detection and your history would all still be sitting on your hardware, and you’d open Frigate in a browser like you did before. That’s not modesty, it’s the architecture: the local system is the product, and everything else is a convenience on top of it. The docs cover the Frigate config and the Tailscale path if you want to walk through steps 3 and 4 this weekend.
The broader point is that “off-grid” isn’t an identity or a lifestyle purchase. It’s four small decisions about where your data sits, and you’re allowed to make two of them and stop.