Aperture vs Frigate: what differs
Frigate and Aperture are both local network video recorders with object detection and no required cloud. They differ in how you install them, what hardware they detect on, their licenses, and how you use them from a phone. This page lists the differences, says when Frigate is the better choice, and says what Aperture does differently. Frigate facts come from its own documentation, read in November 2026.
The short answer
Section titled “The short answer”Choose Frigate if you want the widest choice of detector hardware (a Coral, Hailo, Intel, AMD or Apple chip), an ARM board such as a Raspberry Pi, an MIT license that allows commercial use, face or license plate recognition, or the largest community.
Choose Aperture if you have an x86-64 Linux machine, no accelerator, and want object recognition on the CPU as a supported mode, one installer instead of a container stack, and an iPhone app from the same project. Its license allows noncommercial use only.
Side by side
Section titled “Side by side”| Aperture | Frigate | |
|---|---|---|
| Install | A one-command installer and wizard for Debian or Ubuntu on x86-64. Native systemd services. | A Docker container on any Docker host, or a Home Assistant App. |
| License | PolyForm Noncommercial 1.0.0. Free for noncommercial use. Commercial use needs a separate license. | MIT. |
| Detector hardware | The CPU, or an NVIDIA GPU from Pascal to Blackwell. A Coral and other chips only through an external DeepStack or CodeProject.AI style server. | Hailo, Google Coral, Intel OpenVINO (iGPU, Arc, NPU), NVIDIA GPUs, AMD ROCm, Apple Silicon, Rockchip and community-supported others. |
| Detection on a CPU | A supported mode. The documented time per picture on a desktop CPU is about 9 ms (YOLOX-Nano) to about 300 ms (YOLOX-M). | The documentation says CPU detection should only be used for testing. |
| Recognition beyond objects | People, vehicles, animals, and delivered or picked-up parcels. | Semantic search, face recognition, license plate recognition, bird classification, custom classification and generative AI descriptions. The on-device features download their models on first use. |
| Phones | An iPhone app, plus the web app. | The web interface, with notifications through web push over HTTPS or through Home Assistant. Apps on the App Store that work with Frigate are made by third parties. |
| Home Assistant | Aperture publishes cameras and sensors over MQTT discovery. Nothing to install in Home Assistant. | An official integration installed through HACS. It needs MQTT. |
| Recording | Copies the camera stream without re-encoding. Re-encoding to a lower quality is a per-camera option, on the CPU or an NVIDIA GPU. | Written from the camera stream without re-encoding. |
What Frigate does better
Section titled “What Frigate does better”- More detector hardware. Frigate’s documentation lists detectors for Hailo, Coral, Intel, NVIDIA, AMD, Apple Silicon and Rockchip. Aperture’s own recognition runs on the CPU or an NVIDIA GPU. If your machine has an Intel iGPU, an AMD card or a Hailo module, Frigate can use it and Aperture cannot.
- ARM and small boards. Frigate’s installation page covers the Raspberry Pi and Rockchip boards. Aperture runs on x86-64 only, and its installer stops on anything else.
- A license that allows commercial use. Frigate is MIT. Aperture’s license excludes commercial use unless you get a separate license. See License.
- More kinds of recognition. Face recognition, license plate recognition and semantic search are part of Frigate’s documented features. Aperture does not document these.
- Community and track record. Frigate is widely used, has a large community, and its latest release was 0.18.2 on 15 October 2026. Aperture is new and its community is small. More people means more answers to look up when something breaks.
- Depth of the Home Assistant integration. Frigate is built around Home Assistant, with an official integration and an optional companion card.
What Aperture does differently
Section titled “What Aperture does differently”- Native install. Aperture is installed with one command and runs as systemd services, with no container runtime. The wizard shows every file it writes before it writes anything. See Install.
- Recognition on a CPU is a supported path. The models, their download sizes and their timings per picture are listed on Object recognition. On a CPU with more than two threads, Aperture picks YOLOX-Tiny by default, about 30 ms per picture on a desktop CPU.
- An iPhone app from the same project. It pairs with a code from the web app, and notifications pass through Aperture’s push relay, encrypted so the relay cannot read them. See An NVR with an iPhone app and no cloud.
- Home Assistant connects to Aperture. Aperture is the camera source and publishes through the MQTT broker. See Home Assistant.
Limits to know before choosing Aperture
Section titled “Limits to know before choosing Aperture”- It runs on x86-64 Linux only, and the installer targets Debian and Ubuntu.
- The license is noncommercial. Use inside a business needs a separate license.
- Its own recognition uses the CPU or an NVIDIA GPU. AMD GPUs are detected and not used.
- The community is small, and the iPhone app is version 1.0.
Trying both on the same cameras
Section titled “Trying both on the same cameras”Aperture re-serves every camera over RTSP on port 8554, without re-encoding, so another recorder can read the cameras from it instead of opening a second connection to each camera. Frigate’s documentation lists RTSP re-streaming as a feature, so a Frigate restream can also be added to Aperture as a camera. See The RTSP restream. The restream has no login: allow port 8554 only from machines you trust.
Related
Section titled “Related”Compare all NVRs · Object recognition · Requirements and sizing · Record RTSP cameras on Linux without a Coral