Frigate vs Blue Iris

Frigate is free software under the MIT license. Not freemium, not open-core with the good parts held back. Face recognition and license plate reading landed in 0.16 and cost nothing. The optional Frigate+ subscription is $50 per year and gets you detection models fine-tuned on real security camera footage plus 12 training credits to tune a model on your own cameras (extra trainings are $5 each). You keep any trained models if you cancel. Your real spend is hardware: a used Coral USB runs about $60, a Hailo-8L M.2 module around $70, or you use the Intel iGPU you already have for zero dollars.

Blue Iris 6 is $99.95 for the full license, which covers up to 128 cameras, or $39.95 for the single-camera LE version. That includes one year of updates and support. After that, the software keeps running forever on whatever version you have, and $44.95 per year (auto-renewing) keeps updates flowing. There is a $199.95 priority tier for people who want faster support, but home users don't need it.

Run the five-year math. Frigate: $0 in software, maybe $70 in accelerator hardware. Blue Iris: $99.95 plus about $180 in maintenance if you renew every year, or just $99.95 if you're comfortable freezing on a working version. Neither number should scare anyone, which is exactly why the OS question matters more than the price question.

Frigate 0.17 supports more inference hardware than anything else in this space: Google Coral (USB, mPCIe, M.2), Hailo-8 and 8L (7 to 11 ms inference), OpenVINO on Intel iGPUs, Arc cards, and the NPUs in Core Ultra chips (4 to 35 ms), ONNX on Nvidia and AMD GPUs, TensorRT on Jetson boards, even Apple Silicon as an external detector. Notable 2026 change: the docs now say the Coral is no longer recommended for new installs outside low-power builds. A $150 N100 mini PC running detection on its iGPU via OpenVINO is the new default answer, no dongle required.

Blue Iris spent years leaning on external AI servers, first DeepStack, then CodeProject.AI. Version 6 finally changed that: 6.0.1.2 and later run ONNX YOLOv8 models natively on the CPU or on any DirectX 12 GPU through DirectML. No separate server to install, update, or watch fall over after a Windows update. CodeProject.AI still works if you have a tuned setup with MikeLud's custom models and don't want to touch it.

The practical difference: Frigate treats the accelerator as a first-class config option and squeezes real performance out of $70 hardware. Blue Iris assumes you have a reasonably beefy Windows desktop, ideally with Intel Quick Sync or an Nvidia card, and its sub-stream support keeps CPU load sane even at 20-plus cameras. Both approaches work. Frigate just gives you more ways to do it cheap.

Frigate is a Docker container configured by a YAML file. That sentence either sounded great or terrible, and that's your answer. If it sounded great: docker compose up, point it at your camera RTSP URLs, define detection zones in the browser UI, and your entire NVR config lives in git. The bundled go2rtc handles restreaming, so one connection to each camera feeds detection, 24/7 recording, and WebRTC live view that loads in under a second.

Blue Iris is a classic Windows application, and version 6 finally modernized it: dark mode UI, a rewritten two-file database that survives crashes better, native HTTPS, and hardware-assisted H.265 encode and decode on Intel and Nvidia. Everything is point-and-click. Camera wizards detect most ONVIF hardware, and the UI3 web interface plus the mobile apps are genuinely good for live view and clip scrubbing. Existing v5 licenses carried into v6 without a repurchase, which says good things about how Perspective Software treats customers after two decades of shipping this thing.

Failure modes differ. Frigate's container restarts itself and your config is a text file you can rebuild anywhere in ten minutes. Blue Iris depends on Windows staying healthy: patch reboots, driver updates, and registry-backed config mean you want a dedicated machine that does nothing else, and you should be exporting settings backups.

Frigate has an official Home Assistant integration maintained alongside the project. Camera entities, binary sensors per object type per zone, event snapshots, and a purpose-built Lovelace card all show up automatically. Automations like 'person in driveway after 10 PM, turn on floodlights and announce on the kitchen speaker' are fifteen minutes of work. Everything also publishes over MQTT, so Node-RED and anything else in your stack gets the same events.

Blue Iris can talk MQTT and hit webhooks on alerts, and the community has built solid HA bridges over the years. It works, people run it happily. But it's integration by duct tape compared to Frigate's native support, and every Blue Iris update carries a small chance of your glue needing attention. If Home Assistant is the center of your setup, this section alone should decide it.

If you don't run Home Assistant and never will, flip the weighting. Blue Iris's built-in alerting (push, email, SMS, and now PTZ auto-tracking that follows AI-detected objects in v6) covers most of what home users automate through HA anyway.