Frigate: local security cameras with no cloud
A free recorder that watches your cameras on a computer you own, so no company holds a video feed of your home.
Frigate is a free, open-source network video recorder that stores your IP camera footage at home and runs AI object detection on your own hardware, so a person, a car, or a package triggers an alert without any cloud service involved. For a bitcoiner, the point is physical security without a data trail: the camera feed of your front door and your daily routine stays on your network. As of November 2026 Frigate costs nothing, with an optional $50 a year Frigate+ model subscription.
Why a bitcoiner would run Frigate
The physical threat to bitcoin holders is the wrench attack: someone who thinks you hold bitcoin shows up at your door. My anonymous wallet guide covers how people end up on that list and how to stay off it. This guide covers the other half, which is watching your property without creating a new leak.
A cloud camera sends video of your home, your schedule, and your visitors to a company's servers. That company can be breached, can change its terms, and can answer a subpoena or a request from its own staff. Frigate keeps the whole pipeline on your side. Its own site says all processing is performed locally on your own hardware and your camera feeds never leave your home.
Be realistic about what a camera system does. It gives you a recording, an alert when a person walks up to the door, and evidence afterward. It does not stop a determined attacker, and the recorder itself sits in your house, so someone who walks in can take it. Cameras are one layer, next to not advertising your holdings, keeping your bitcoin on a hardware wallet, and keeping its seed phrase backed up where nobody can find it.
What Frigate does
Frigate is an MIT-licensed project, and its repository is the source for everything below. It describes itself as a complete and local NVR designed for Home Assistant with AI object detection that uses very low overhead motion detection to decide where to run detection, so it only looks for objects when and where it needs to.
- Recording. Frigate writes recordings straight from the camera stream without re-encoding, and each camera gets a retention policy. You can keep everything for a few days, motion-only footage for longer, and clips with alerts or detections longest.
- Object detection. A detector model labels what moved. By default person is the only tracked object, and you can add cars, dogs, and others from the label list.
- Zones. A zone is an area of the camera frame you draw, such as the front steps or the driveway. Frigate checks whether an object is inside using the bottom center of its bounding box, so you can alert on a person on the steps and ignore one walking past on the sidewalk.
- Alerts. Frigate sends web push notifications to a browser, and it talks MQTT, which is how it connects to Home Assistant for the rest of your automations.
What you need
Frigate's hardware page is the reference. The short version is cameras, a small Linux server, and a detector.
| Part | What the docs say | My pick |
|---|---|---|
| Cameras | IP cameras. H.264 video and AAC audio are the most compatible, multiple substreams help, and Wi-Fi cameras are not recommended because their streams are less reliable and lose connection. The docs recommend Dahua, Hikvision, and Amcrest, in that order. | Wired PoE cameras with a separate low-resolution substream for detection |
| Server | An Intel CPU with AVX and AVX2 that can run Debian. The docs' current favorite is a Beelink EQ13, an N100 mini PC with dual network ports, which lets you build a private camera network that cannot reach the internet. | A small x86 box with two network ports, the same kind I describe in my home server guide |
| Detector | Optional but strongly recommended. Supported hardware includes Hailo-8 and Hailo-8L, Intel OpenVINO on integrated GPUs and NPUs, Nvidia GPUs, AMD ROCm, and Apple Silicon. CPU detection is for testing only. The Coral is no longer recommended for new installs, except where power use must be very low or the hardware cannot use another accelerator. | An Intel iGPU or NPU through OpenVINO, so there is no extra card to buy |
| Storage | Any disk you map to the recordings folder, including an external USB drive. Frigate manages the folder layout itself, so do not browse or reorganize it by hand. | A dedicated disk for video, kept apart from anything you care about |
Frigate does not give a storage-per-camera figure, because bitrate depends on the camera. Set the bitrate in the camera's own settings, check how much a day uses, and size the disk from that.
Installing Frigate
Frigate's installation docs call Docker Compose the recommended method and say it runs best on Docker installed on bare-metal Debian-based Linux. They do not recommend a virtual machine on Proxmox or ESXi, though some users make it work. For the current commands and the compose file generator, follow those docs.
- 1Install Docker on a Debian-based Linux machine, and set aside a folder for recordings.
- 2Generate your compose file with the generator on the installation page, which asks for your hardware and storage paths.
- 3Start Frigate. On first start it generates an admin password and prints it in the logs. Log in and replace it with your own.
- 4Add each camera by its stream address, as in the getting started guide. Give the low-resolution substream the detect role and the high-resolution stream the record role.
- 5Draw a zone for the area you care about and name the objects it should watch for.
- 6Watch the first few days of alerts and fix the false ones with zones and object filter masks before you rely on it. The docs say motion masks do not reduce false positives.
If you already run Home Assistant OS, you can install Frigate as a Home Assistant app from the frigate-hass-addons repository. The docs warn that Nvidia GPUs, AMD GPUs, and Intel NPUs are not supported there, so check your detector first. My Home Assistant guide covers the base setup.
On umbrelOS 0.5 or later, Frigate is in the Umbrel app store with cameras configured inside the app. It shares a box with your node, so an Umbrel machine can do both jobs. The "Frigate" in Start9's community registry is a different project, an Electrum server from the Sparrow Wallet team, so it does not record cameras. On StartOS, run Frigate on a separate Docker host.
Keep the cameras off the internet
A local NVR defeats its own purpose if the cameras still phone home. Treat every camera as untrusted hardware, whatever the brand, and cut its route to the internet.
- Put the cameras on their own network. A separate VLAN or a second network port on the server works, with the router blocking all outbound traffic from that network. The docs' dual-port server pick exists for exactly this reason.
- Allow the cameras to talk to the Frigate server and nothing else. Frigate pulls the streams, so the cameras never need to reach out.
- Do not port forward anything. Frigate's port 8971 is the authenticated interface, while port 5000 is unauthenticated and the docs say to limit access to it. The RTSP restream on port 8554 is also unauthenticated by default.
- For viewing away from home, join your phone and the server to a private VPN such as Tailscale instead of opening a port. My Tailscale vs WireGuard guide compares the two.
- Know Frigate's own outbound traffic. The network requirements page says core functions need no internet after initial model downloads, and lists what does connect: a version check you can turn off, push notifications through the browser vendor's push service, Frigate+ if you add an API key, any cloud Generative AI provider you configure, and a fallback to Google's public STUN server in the web interface's WebRTC player.
Notifications are the one real trade-off. The notification docs say Frigate must be accessible externally for alerts to work away from home, so test that setup with your VPN before you trust it.
What Frigate costs
Frigate itself is free. You pay for hardware and for electricity. The optional Frigate+ subscription is $50 a year as of November 2026. It gives you access to models trained on real security camera images, and up to 12 fine-tuned models a year, with extra fine-tunes at $5 each.
You do not need it. Face and license plate recognition are built into Frigate and work with or without Frigate+, and you do not have to upload your own images. If you cancel, models trained during your subscription stay downloadable. The privacy catch is that Frigate+ talks to api.frigate.video to download models and, if you use the training features, to upload snapshots, so I would skip it unless you want those extra labels.
Frequently asked questions
Is Frigate free?
Yes. Frigate is open source under the MIT license and has no required fees. The only paid part is the optional Frigate+ models subscription at $50 a year, which most home setups can skip.
Does Frigate need the internet?
Not for core use. After its models are downloaded the first time, recording, detection, and live view run locally. Push notifications, Frigate+, and cloud Generative AI providers do use the internet, and a version check can be disabled.
Do I need a Google Coral to run Frigate?
No. The docs no longer recommend a Coral for new installs, outside very low-power setups or hardware that cannot use another accelerator, and list Hailo, Intel OpenVINO, Nvidia, AMD, and Apple Silicon detectors as alternatives. Running detection on the CPU alone is for testing only.
Can I watch my Frigate cameras from outside my house?
Yes. Connect your phone to your home network through a VPN such as Tailscale and open Frigate's authenticated interface. Avoid port forwarding, which exposes the login page to the whole internet.
Will security cameras protect me from a wrench attack?
They give you alerts and a recording, but they do not stop an attacker, so treat them as one layer. Not being identified as a bitcoin holder matters more, and the anonymous wallet guide covers that.