What Is Frigate? A Complete Guide to Local AI Security Cameras with Home Assistant

Posted on by

Frigate is an open-source network video recorder that runs on your own hardware and uses local AI to understand what your security cameras are seeing.

Instead of treating every pixel change as important motion, Frigate can identify people, vehicles and other objects, organize footage around what actually happened and make that information available to Home Assistant.

That last part is what makes Frigate particularly interesting to me.

A traditional NVR records cameras.

Frigate turns cameras into sensors.

Once Home Assistant knows that a person entered the driveway, a vehicle arrived in a particular zone or someone approached the front porch, video becomes another source of information the house can use for automation — and most of that processing can stay inside your own network.

Frigate in One Sentence

Frigate is a local NVR that combines IP-camera recording, object detection and home automation so your cameras can tell you what happened instead of simply giving you hours of video to watch.

Frigate on Home Gadget Geeks

Frigate has come up several times on Home Gadget Geeks, but my conversation with Mike Wieger in HGG689 was what convinced me it deserved a guide of its own.

Mike has experimented with Blue Iris, Sighthound and other camera platforms, and he walked me through why Frigate 0.18 had become his preferred local NVR — an easier web interface, smarter retention, local object detection, face recognition and Home Assistant integration.

Gavin Campbell gave me another useful perspective in HGG690. He has used Frigate and Blue Iris but currently values the integrated experience he gets from UniFi Protect.

That matters because Frigate is not automatically the best answer for every house. This guide is about figuring out when it is.

Why Frigate 0.18 Matters

Frigate has been powerful for a long time. What has changed is how much of that power is now accessible without living inside configuration files.

Frigate 0.18 brought much more configuration management into the web interface. Cameras, recording, motion, detectors, review settings and other parts of the system are easier to manage without hand-editing YAML.

The underlying platform did not suddenly become capable.

It became easier to use.

What Does Frigate Actually Do?

At its simplest, Frigate accepts video streams from network cameras and records them like an NVR. But recording is only one layer.

  • detect motion
  • identify objects
  • track where those objects move
  • organize activity into alerts and detections
  • retain important footage longer
  • recognize known faces
  • read license plates
  • search video semantically
  • expose camera activity to Home Assistant

The goal is no longer simply record everything.

It becomes understand what happened and keep the parts that matter.

Why Local Processing Matters

Security cameras may capture some of the most private data in your home: when people arrive and leave, what vehicles are present, activity around the property and sometimes activity inside the house.

Frigate is designed around local processing. Your camera streams can stay on your network, your recordings can stay on storage you control and object detection can run on hardware inside your home.

Optional Internet-connected features exist, but cloud processing is not required for Frigate to function as a useful NVR.

Watch: I Stopped Using Cloud Cameras. Here’s Why. This short clip explains why keeping camera video local became such an important part of my approach to Frigate.

For me, that is one of its biggest advantages. A camera system should not have to send every frame outside my house just to tell me somebody is standing in the driveway.

What Cameras Work With Frigate?

Frigate works with many IP cameras, but not every camera is equally good for the job. If I were building a Frigate system from scratch, I would look for four things first:

  • Power over Ethernet
  • H.264 video
  • AAC audio
  • multiple video streams

Why PoE Cameras Are Preferred for Frigate

Power over Ethernet lets one Ethernet cable provide both network connectivity and power.

For a permanent camera installation, I prefer that to Wi-Fi whenever running cable is practical. Security video is a continuous network workload, and a wired connection removes one more source of instability.

Wi-Fi cameras can work. I simply would not choose Wi-Fi first for a new multi-camera Frigate system.

Multiple Streams Matter More Than Megapixels

A good IP camera can provide a high-resolution stream for recording and a smaller stream for detection.

Recording stream: 2592 × 1944 at 15 fps
Detection stream: 1280 × 720 at 5 fps

Frigate can analyze the lighter stream while preserving the high-resolution stream for recordings, reducing processing requirements without sacrificing the quality of the footage you keep.

This is why I would not choose a camera solely because the box says 4K, 8MP or 12MP. Image sensor size, lens quality, night performance and placement may matter more than raw resolution.

The real question is: Can I identify what I need to identify?

Which Camera Brands Work Well?

Frigate’s current hardware guidance recommends Dahua, Hikvision and Amcrest camera families. Those products commonly expose the kinds of features Frigate needs: RTSP, ONVIF, multiple streams, adjustable resolution, configurable frame rates and configurable keyframe intervals.

For many home users in the United States, Amcrest is an approachable starting point because the cameras are widely available and their network streams are well documented. I also put together a dedicated guide to the best cameras for Frigate if you want specific PoE, low-light, panoramic, Reolink and PTZ recommendations.

That does not mean you should replace cameras you already own. If an existing camera exposes a stable network stream, test it first.

What About Reolink?

Reolink cameras are popular and affordable, and Frigate supports them, though configuration can vary by model and generation.

If I were buying specifically for Frigate, I would favor a model with an established Frigate configuration rather than simply buying the camera with the largest resolution number.

A Good First Frigate Camera

For a first project, I like the idea of a fixed PoE turret camera. It gives you reliable Ethernet, no batteries, no Wi-Fi dependency, separate recording and detection streams, night vision, weather resistance and straightforward RTSP connectivity.

Frigate currently includes the Amcrest IP5M-T1179EW-AI-V3 among its recommended camera options.

Affiliate disclosure: Some links in this guide are Amazon affiliate links. If you purchase through them, I may earn a commission at no additional cost to you.

Amcrest IP5M-T1179EW-AI-V3 5MP PoE Turret Camera on Amazon

I would still verify the exact model number before buying. Amazon listings sometimes combine similar model names or hardware revisions.

A Small PoE Switch for a Few Cameras

If you are building a small two-to-four-camera system and do not already have PoE available, you do not need a huge managed switch.

The TP-Link TL-SG105PE provides five Gigabit Ethernet ports, four of them PoE+, with a 65-watt PoE budget. That is a sensible size for a small Frigate installation.

TP-Link TL-SG105PE 5-Port Gigabit PoE+ Switch on Amazon

Adding Your First Camera

Older Frigate tutorials can make the platform look more intimidating than it is today.

Historically, adding a camera often meant finding the right RTSP URL, editing configuration files and restarting Frigate until everything worked.

Manual configuration still exists, but newer Frigate versions also provide camera-management tools in the web interface.

Frigate’s current Add Camera Wizard can probe compatible cameras, test available streams, show their resolution, frame rate and codecs, assign separate detection and recording roles, configure go2rtc and warn about common stream problems.

What Is ONVIF?

ONVIF is a standard that lets compatible security devices describe their streams and capabilities to other systems. You do not need to understand the whole standard. For Frigate, the useful part is that ONVIF can make camera discovery and configuration considerably easier.

What Is go2rtc?

You will probably see the name go2rtc while configuring Frigate. It is the streaming layer Frigate uses for restreaming camera feeds, live viewing and some integrations.

Instead of every application opening another direct connection to a camera, go2rtc can connect once and make the stream available where it is needed. It also plays an important role in WebRTC live viewing and, with supported cameras, two-way audio.

Frigate handles recording and intelligence. go2rtc helps move the video efficiently.

Can Frigate Control PTZ Cameras?

Yes. Frigate can control compatible ONVIF pan-tilt-zoom cameras, and some cameras can automatically track moving objects as they cross the image.

But basic PTZ support does not guarantee autotracking. Frigate’s autotracking requires specific ONVIF movement capabilities, and camera firmware implementations vary widely.

Does Frigate Need 30 FPS for Detection?

Usually not. Smooth video playback and object detection are different jobs.

For many normal security-camera situations, around 5 frames per second can be enough for detection. Frigate does not need 30 slightly different images every second to decide that somebody walked into the driveway.

There are exceptions, especially when something moves quickly through a narrow area, but the larger principle holds:

The best Frigate configuration is not the one with every number turned to maximum.

It is the one that gives each part of the system exactly what it needs.

What Hardware Does Frigate Need?

Frigate hardware advice gets confusing because Frigate performs several different compute jobs. I think about them in three buckets:

  1. Video decoding
  2. Object detection
  3. AI enrichments

Video Decoding

Every camera sends compressed video, and before Frigate can analyze or display it, that video must be decoded.

A CPU can do the work, but modern CPUs and GPUs often contain dedicated video hardware that does it far more efficiently.

For many home installations, Intel integrated graphics are particularly attractive. An inexpensive business mini PC may not look impressive by gaming-PC standards, but its media engine can be very good at exactly the type of work Frigate needs.

Object Detection — Do You Still Need a Google Coral?

Object detection answers: What is in this image?

Google Coral became one of the most popular Frigate accelerators for years, but it is not a requirement.

Frigate now supports several detector paths, including Intel OpenVINO, NVIDIA, AMD, Hailo, Google Coral and other supported hardware.

So no, you do not still need a Coral. A better question is: What capable hardware do I already have?

For example, OpenVINO supports Intel platforms going back to 6th-generation Skylake systems, depending on the device and configuration.

Why Intel Is Interesting

For a typical home Frigate build, Intel hardware can offer a particularly simple architecture. One small system may be able to run Frigate, decode camera video in hardware, perform OpenVINO object detection, run Docker, communicate with Home Assistant and handle several cameras at modest power.

I am deliberately not linking a specific mini PC here. Model availability and hardware revisions move quickly, and I would rather leave this slot unmonetized than recommend a machine whose current Amazon listing I cannot confidently verify.

AI Enrichments

Frigate can go beyond basic object detection with features such as face recognition, semantic search and license-plate recognition.

Those workloads are separate from normal detection. An accelerator that is excellent at answering Is this a person? may not be the hardware accelerating the model used to answer Which person is this?

A Practical Hardware Ladder

Stage 1: Reuse What You Own

Good for learning, a few cameras and basic detection. Cost could be $0.

Stage 2: Dedicated Intel Mini PC

A strong option for several cameras, low power use, hardware decoding and OpenVINO.

Stage 3: Dedicated AI Accelerator

Add one when you can identify the detection workload it will improve.

Stage 4: GPU-Accelerated Server

More appropriate when the workload grows into many streams or heavier enrichment models. At that point, the machine is becoming a local video-AI server, not simply an NVR.

How Much Storage Does Frigate Need?

There is no useful universal answer such as four cameras need a 4 TB drive.

Storage depends on bitrate, codec, camera count and retention policy.

Storage per day ≈ bitrate in Mbps × 10.8 GB

  • 1 Mbps ≈ 10.8 GB/day
  • 2 Mbps ≈ 21.6 GB/day
  • 4 Mbps ≈ 43.2 GB/day
  • 8 Mbps ≈ 86.4 GB/day

Four cameras averaging 4 Mbps could generate roughly 173 GB per day. Thirty days of continuous recording would exceed 5 TB.

You Do Not Need to Keep Everything Forever

Frigate can apply different retention policies to different types of footage. For example, all video for 3 days, motion footage for 7 days and important alerts and detections for 30 days.

Those are examples, not recommended universal settings.

Retention is a policy decision, not simply a hard-drive-size decision.

SSD or Surveillance HDD?

An SSD can make sense for the operating system, containers and application data. For large amounts of continuous recording, surveillance hard drives provide a lot of capacity for the money.

For many systems, I would prefer an SSD for the host and a surveillance HDD for recordings. Common surveillance-drive families include WD Purple and Seagate SkyHawk.

I am leaving this section without an Amazon link for now. A technically appropriate drive is easy to identify; a clean, currently available Amazon listing with a reliable buy box was not. That is not a reason to force an affiliate link into the article.

Does Frigate Re-Encode Camera Recordings?

Normally, no. Frigate writes recordings directly from the camera stream without re-encoding them.

Export Important Footage

If something important happens, export it. Do not increase whole-system retention dramatically because there is one event you want to preserve.

How Does Frigate Work With Home Assistant?

This is where Frigate becomes much more interesting to me.

A normal NVR records video. Frigate can tell Home Assistant what is happening in the video.

Instead of motion detected, the house can know a person entered the driveway.

Does Frigate Require Home Assistant?

No. Frigate works independently as an NVR. Home Assistant becomes valuable when you want camera events to interact with the rest of the house.

Why MQTT Matters

For the official Frigate Home Assistant integration, MQTT needs to be configured, and Frigate and Home Assistant should connect to the same MQTT broker.

MQTT acts like a lightweight message bus for much of the state and event information Frigate exposes. The integration also uses Frigate’s own API and media interfaces, so MQTT is not the entire connection between the two systems.

Zones Make Camera Automation Better

Imagine a front camera that sees the street, sidewalk, lawn, driveway and porch. You probably do not want the same automation whenever a person appears anywhere in the image.

Frigate lets you create zones such as sidewalk, driveway and front_porch.

A Better Front-Door Automation

Someone walks down the sidewalk. Nothing happens.

They turn into the driveway. Home Assistant could turn on exterior lights, send a notification and display the camera.

They enter the porch zone. A higher-priority automation might send a snapshot, brighten the porch light and display the camera on an indoor screen.

The camera did not change. The intelligence around it did.

The Best Notification Is Not Every Notification

A security system that sends too many alerts eventually becomes a system nobody pays attention to.

The goal should not be tell me whenever anything happens. It should be tell me when something happens that I actually care about.

Frigate Gives Home Assistant Eyes

This is the core idea for me.

ESPHome lets me build sensors Home Assistant did not have before. Frigate does something similar with cameras.

It turns video into information the rest of the smart home can understand.

Frigate gives Home Assistant eyes.

Where Does Frigate Run?

Frigate is software. You provide the host. That could be an Intel mini PC, a Linux server, an Unraid server, another Docker host or a supported Home Assistant environment.

Docker Compose is one of the primary deployment methods.

Frigate on Unraid

Unraid is a natural Frigate host because it already provides Docker, storage and hardware pass-through. If you already run Unraid, I would test Frigate there before assuming you need another computer.

Hardware Pass-Through Matters

The Frigate container needs access to whichever hardware you expect it to use — Intel graphics, an NVIDIA GPU, Hailo, Coral or another supported accelerator.

Frigate App vs. Home Assistant Integration

The Frigate App runs Frigate. The Frigate integration connects Frigate to Home Assistant.

Remote Access

The authenticated Frigate interface currently uses port 8971. Frigate also has an internal unauthenticated interface on port 5000.

Port 5000 should be treated as an internal trusted interface, not something casually exposed to the Internet.

Face Recognition

Frigate can recognize known people locally. Person detection happens first, and Frigate can then examine a face and compare it with people you have trained into the system.

I would still treat recognition as a useful signal rather than an infallible identity system.

License Plate Recognition

Frigate can recognize license plates associated with vehicle detections.

Camera placement matters enormously. A great general-purpose camera can still be terrible at reading plates. If LPR really matters, design the camera position around that job.

Semantic Search

Semantic search may be one of Frigate’s most interesting newer capabilities. Instead of searching only for labels such as person, car or dog, you can search recorded activity based on meaning.

Examples might include a person wearing a red shirt, a delivery truck, a person carrying a package or a blue vehicle.

What Is Frigate+?

Frigate itself is open source. Frigate+ is an optional paid service focused on improved object-detection models.

Users can submit and verify selected images from their own cameras, including true and false detections. Those submitted examples can be used as part of the Frigate+ model-training and fine-tuning process.

Mike gave me a good example on HGG689: his system repeatedly confused his dog with a cat, and Frigate+ gave him a workflow for showing the system which detections were correct and which were not.

Does Frigate+ Upload All Your Video?

No. Using Frigate+ does not mean your live camera feeds are continuously sent to the cloud for analysis. The detection model runs locally. Images are uploaded when you intentionally submit them to the Frigate+ workflow.

Do You Need Frigate+?

No. I would start without it.

Is Frigate Private?

Frigate’s local architecture gives you more control. It also gives you more responsibility.

Cameras deserve stronger security than many ordinary smart-home devices. I would pay attention to strong passwords, camera firmware, secure remote access, network segmentation where appropriate, limiting administrator access and disabling camera cloud services you do not use.

A temperature sensor cannot tell someone when your family arrived home. A camera might. Treat camera infrastructure accordingly.

Frigate vs. Blue Iris vs. UniFi Protect

Frigate is not the only good option.

Blue Iris is a mature Windows-based NVR. UniFi Protect provides a much more integrated camera and recording ecosystem.

Frigate emphasizes local AI, flexible camera support, Docker, Home Assistant, automation and hardware choice. That flexibility is one of its biggest strengths. It is also part of the complexity.

Mike Wieger’s experience has pushed him toward Frigate. Gavin Campbell’s experience has led him toward UniFi Protect. Both perspectives are useful.

I also published a separate full comparison of Frigate vs. Blue Iris vs. UniFi Protect. That decision deserves its own article rather than turning this guide into three guides at once.

Who Is Frigate Best For?

  • want local camera processing
  • already use Home Assistant
  • are comfortable with Docker
  • want to choose your own cameras
  • care about object detection and zones
  • want deep automation
  • already run a homelab
  • want control over retention
  • prefer less cloud dependence

Who Might Prefer Something Else?

Frigate may not be ideal if your priority is I want to plug it in and never think about it again.

Someone who wants a tightly integrated appliance may prefer UniFi Protect. Someone deeply invested in Windows may prefer Blue Iris.

Frequently Asked Questions

Is Frigate free?

Yes. The core Frigate software is open source. Frigate+ is optional.

Does Frigate require the Internet?

Not for normal local NVR operation once the required software and models are present. Internet access may be needed for software updates, model downloads and optional connected services such as Frigate+.

Does Frigate record 24/7?

It can. Continuous recording is optional, and different categories of footage can have different retention periods.

Does Frigate work without Home Assistant?

Yes. Frigate is a standalone NVR with its own interface. Home Assistant is optional.

Does Frigate need MQTT?

Not for standalone NVR operation. The official Home Assistant integration does require MQTT to be configured, and many entities depend on Frigate and Home Assistant sharing the same MQTT broker.

Does Frigate require a Google Coral?

No. Coral is one supported detector option among several.

Can Frigate run on Unraid?

Yes. Unraid is a natural environment for Frigate because it combines Docker, storage and hardware pass-through.

Can Frigate use Wi-Fi cameras?

Yes. I still prefer wired PoE for permanent cameras when practical.

Can Frigate do two-way audio?

With supported cameras and the appropriate go2rtc configuration, it can. Compatibility depends on the camera and its audio backchannel implementation.

Does Frigate have AI chat?

Frigate 0.18 includes optional GenAI capabilities when compatible providers are configured. That is separate from Frigate’s normal local recording and object-detection functions. You do not need an LLM to use Frigate.

How many cameras can Frigate handle?

There is no universal number. Capacity depends on stream resolution, frame rate, video decoding hardware, object detector, enabled AI enrichments, camera activity and storage performance.

My Frigate Starter Architecture

If I were building a straightforward four-camera Frigate system today, I would start with:

  • four wired PoE cameras
  • multiple camera streams
  • H.264 video
  • a small Gigabit PoE switch
  • Intel-based Frigate host
  • 16 GB RAM
  • hardware video decoding
  • OpenVINO object detection
  • dedicated recording storage
  • Docker Compose or Unraid
  • MQTT
  • Home Assistant

I would not buy every accelerator available. I would get the system working first, then measure it, then upgrade the actual bottleneck.

Where Should You Start?

Start with one camera. Get Frigate running and make sure you can:

  1. View the camera.
  2. Record it.
  3. Detect a person.
  4. Create a useful zone.
  5. Find the event afterward.

Then connect Home Assistant and build one automation that solves a real problem.

For example: When a person enters the front porch zone after dark, turn on the porch light.

That one automation demonstrates almost everything that makes Frigate interesting.

The camera sees the world.

Frigate understands what the camera sees.

Home Assistant decides what the house should do about it.

Frigate gives Home Assistant eyes.