Frigate 0.17 FAQ

[hawkeye217]

This FAQ is maintained by the Frigate developers to help answer some of the common questions with Frigate 0.17. If you have a question that's not answered here, be sure to read the release notes, the documentation, and search the recent discussions first. If you still need help, open a support discussion.

Some questions may also be answered in the older version, Frigate 0.16 FAQ, so be sure to read those as well.

Updated March 5, 2026

Upgrade issues

1. I just upgraded my Home Assistant add-on and now my Coral TPU is not being detected.
   The add-on's protection mode switch may be re-enabled by Home Assistant after upgrading. You need to disable protection mode for the add-on or use the standard add-on (not "Frigate Full Access").

2. I'm seeing Error parsing jwt: bad_signature: in my logs after upgrading.
   You should reload the Frigate Home Assistant integration.

3. Frigate is crashing on startup or when I try to train a classification model with an illegal instruction error.
   Frigate 0.17 requires a modern CPU with AVX (Advanced Vector Extensions) instructions. This error typically happens when:

   • Your CPU lacks AVX instructions required by TensorFlow
   • You're running Frigate in a virtualized environment where the CPU type is incorrectly configured

   If you're running Frigate in Proxmox (or another virtualization platform), change the CPU type setting from a specific model (like x86-64-v2-AES) to "host" mode. This passes through the actual CPU capabilities to the container rather than emulating a specific CPU model.
   In rare cases, server platforms may also have a BIOS setting to toggle AVX, assuming the CPU model already supports AVX instructions.

4. Frigate is hanging on startup.
   As the breaking changes in the release notes indicate, the auto detection logic for camera resolution has changed. Some cameras fail to correctly advertise their resolution, and in previous versions, a default value was assumed that was not always correct. You may need to explicitly define detect resolution width and height for cameras in your config if Frigate hangs on startup.

Usage

1. How can I use classification in my Home Assistant automations?
   The official Frigate HA integration automatically creates sensors for each classification model you define, which can be used in automations. Also, sub labels added to objects through object classification will already work with the sgtbatten notification blueprint or other third-party solutions like frigate-notify. An alternative to the official integration would be to use MQTT.

2. I'm using the new classification wizard, and none of the images show the correct state. What should I do?
   You do not need to train every state or class during the initial wizard. The wizard is designed to get a model working quickly using the examples that are readily available, not to exhaustively label your entire recording history. Adding the ability to browse back through old recordings just to find rare or missing states would add significant UI/UX complexity for limited benefit. In practice, models often perform well even when some classes have only a small number of examples. As Frigate continues running, any previously missing states or classes will appear automatically in the Recent Classifications view, where you can label them and retrain the model as needed.

3. Where is the Debug view? It's not in Settings anymore.
   This was covered in the release notes. Access the Debug view from single camera Live view by pressing the cog icon and turning on the Debug switch. You can also access Debug view directly from the Live dashboard by right clicking/long pressing on a camera and selecting Debug from the context menu.

4. Can I set up the new classification models feature to detect new objects?
   No. The new classification models feature does not allow you to add or train entirely new detectable or tracked object types. Object classification models only run after Frigate has already detected an object, and state classification models run on a fixed area of the camera frame, independent of object detection. They can label the state of something (e.g., door open/closed) or classify within an existing object type (e.g., dog → Fido), but they cannot introduce new objects to the detection pipeline.

5. Frigate loses connection to my cameras and then consumes a lot of memory until it becomes unresponsive. Why?
   Users with this issue are often running WiFi cameras, which are not recommended. A number of users have also reported that this issue can be mitigated by using input args as preset-rtsp-generic instead of preset-rtsp-restream. But in general, this is usually a result of packet loss from WiFi cameras and is exacerbated by ffmpeg trying to recover from it.

6. I'm having trouble using the mask/zone editor and the classification selection square, I can't click points or move the square.
   This is due to Brave Browser's underlying extreme anti-fingerprinting code. To use Brave Browser with Frigate, you should disable Brave Shields.