A motion sensor knows something moved. A camera can know what it was: you or a stranger, your car or the courier's van, a person or the cat. That is a much better trigger for a smart home than "motion in the hall".

Agent DVR does the recognizing on your own computer and passes the result on. Your cameras show up in Home Assistant through its own Agent DVR integration, with no YAML to write, a recognized face or number plate can publish an MQTT message or call a URL, and everything Agent DVR can do is available through an HTTP API. It runs on your own network, with no cloud in between.

It knows who has arrived

Face recognition and number plate recognition are built in and run locally. Enrol the faces and plates that matter, and each one becomes a trigger: an action runs when that person or that car is seen, and it can tell them apart from everyone else. People use it to:

open the gate for a known plate open the garage door turn on the lights when you get home announce who is at the door disarm for family alert on a stranger

It plugs into what you already run

Home Assistant has an Agent DVR integration in its own list: add it with your server's address and you get every camera, an alarm panel to arm and disarm, and actions to record, take a photo and switch alerts. Turn on MQTT discovery as well and each camera adds motion, alert, recording and online sensors. It also works with:

Home Assistant MQTT Amazon Alexa IFTTT ioBroker webhooks an HTTP API AI assistants

It hears the house too

Audio recognition listens to a microphone or a camera's audio and names what it hears, from a list of more than 500 sounds. Any of them can trigger an automation, including ones no other sensor in the house would notice:

the doorbell a knock a smoke alarm breaking glass a baby crying a dog barking

Home, Away and Night

Profiles switch the whole system between set-ups in one step: which cameras record, which alert, how sensitive they are. Change profile however suits the moment:

by voice with Alexa on a schedule from Home Assistant over MQTT from the API by asking an assistant

What it looks like

Who came to the house today?

Five recordings on the driveway and front door cameras today. Looking at the frames and at what was recognized:

  • 08:05 - your car left. Plate recognized.
  • 11:20 - a courier's van stopped for a minute. Plate not on your list.
  • 15:40 - Sam at the front door. Face recognized, and the hall light rule ran.
  • 17:55 - your car came back. Plate recognized, and the garage door rule ran.

The fifth was the cat on the doorstep. Nothing unexpected.

Things to ask

  • Who came to the front door today?
  • Arm the system with the Away profile.
  • Is the garage door open? Look at the garage camera.
  • What time did the car come back?
  • Which faces do you recognize?

Rules to set up by asking

  • When the driveway camera sees my number plate, call my garage door webhook.
  • When Sam is at the front door, turn on the hall light.
  • If there is a person at the door after dark who is not one of us, tell me.
  • Arm the system at 11pm every night.

Rules run on your own server with no assistant connected. They need Agent DVR 8.0.7 or later, and questions about what is in the picture need an AI provider set up in Server Settings - AI Settings. More rule examples

Why build it this way

A better trigger than motion

Automations that fire on "a person", "this face" or "this plate" do the right thing far more often than ones that fire on movement, and they do not run for the cat.

Home Assistant, the official way

The Agent DVR integration ships with Home Assistant: add it from the integrations list, enter the server address, and your cameras and the alarm panel are there, with no YAML to write. A camera you add later appears too.

Voice, when your hands are full

With the Alexa skill you can say "Alexa, arm Agent in home mode" or "Alexa, show Agent camera" on an Echo Show. Or ask an AI assistant in your own words.

Local first

Recognition, recording, MQTT, Home Assistant and the API all run on your own network. Nothing has to leave the house for the gate to open.

What you need

1

A camera

Almost any IP, ONVIF, RTSP or USB camera works, including a spare webcam. You do not need a special bird, pet or doorbell camera. Look yours up in the camera database.

2

Agent DVR on a computer

Install it on Windows, macOS, Linux, Docker or a Raspberry Pi. It records to your own disk and is free for private use on your own network. Download.

3

An AI assistant

Connect Claude, ChatGPT, Gemini CLI or any other MCP client and start asking. Free on your own network. From your phone or the cloud it is part of an iSpyConnect subscription from $7.95 a month. How it works.

Set it up

1

Connect Home Assistant

Add the Agent DVR integration in Home Assistant and enter the server address. For the extra sensors, turn on MQTT discovery too.

2

Teach it who is who

Turn on face recognition or number plate recognition on the cameras that see the door and the drive, and enrol the faces and plates that matter.

3

Wire up the result

Add an action on Face Recognized or License Plate Recognized that publishes to MQTT or calls a Home Assistant webhook, or ask an assistant to set it up.

Good to know: Face and number plate recognition need a camera that sees the face or the plate clearly, close up and well lit: a wide-angle camera across a driveway will read few plates at night. A camera can be fooled by a photograph, so do not let a face or a plate alone unlock anything that matters; use it for convenience, such as lights or a garage door you can see. Audio recognition names sounds it has learned from everyday recordings, so treat a sound as a prompt to look, not as proof, and it needs an Agent DVR license or a subscription. The Home Assistant integration needs only the server's address; the extra sensors need an MQTT broker, usually the Mosquitto add-on. Alexa and IFTTT are linked through an iSpyConnect account and need a subscription. Rules that an assistant makes can call a URL or publish to MQTT, so it switches a light by calling Home Assistant or the light's own API.

Frequently asked questions

Does Agent DVR work with Home Assistant?

Yes, in two ways that work together. The Agent DVR integration that ships with Home Assistant adds your cameras, an alarm panel to arm and disarm the system, and actions to start and stop recording, take a photo and switch alerts on and off. MQTT discovery adds motion, alert, recording and online sensors for each camera and microphone. Neither needs any YAML, and you can also embed the Agent DVR view in a dashboard. See Home Assistant in the user guide.

Can a recognized face or number plate trigger an automation?

Yes. Add an action on Face Recognized or License Plate Recognized, set the name or the plate it should match, and choose what it does: publish an MQTT message, call a URL such as a Home Assistant webhook, speak through a camera, or run several tasks in turn. An AI assistant can set the same thing up from a sentence.

Can I control it by voice?

Yes. The Alexa skill arms and disarms Agent DVR in its Home, Away and Night profiles and shows a camera on an Echo Show. Any AI assistant with a voice mode, such as the Claude or ChatGPT app, can do more: ask what happened, look at a camera, arm the system or set up a rule.

Is there an API?

Yes. Everything in Agent DVR can be driven over a local HTTP API, and commands can also be sent over MQTT. Actions can call out to any URL with the event's details in the request. See the API documentation.

What does it cost?

Agent DVR is free for private use on your own network, including Home Assistant, MQTT and the API. Alexa, IFTTT, the mobile apps and access away from home are part of an iSpyConnect subscription from $7.95 a month, one flat price for all your cameras. The AI provider you use for the assistant may have its own costs.

Running Frigate with Home Assistant today? See Agent DVR vs Frigate, or Agent DVR vs Blue Iris.

Download Agent DVR How the AI assistant works

More things your cameras can do

Security analytics

Loitering, left bags, covered cameras and line crossing on any camera, with PTZ tracking, speed and sound alerts, and tools to pull the evidence together afterwards.

See how it works

Package delivery alerts

Find out when a package is delivered, who came to the door while you were out, and whether the package is still on the step.

See how it works

AI pet camera

Ask what the dog did while you were out, look in live from your phone, and get told when the cat is on the kitchen counter.

See how it works

AI bird feeder camera

Point any camera at the feeder and ask which birds visited, when the feeder is busiest and whether the squirrel is back.

See how it works

Checking in on a relative

See that Mum or Dad is up and about, find out whether the carer came, and get told when something needs a look, without phoning every hour.

See how it works

Barn and foaling camera

Watch the foaling box, calving pen or chicken coop from your phone, ask what happened overnight, and get an alert when something needs a look.

See how it works

Machine monitoring

Get told when an alarm sounds, a warning light comes on, or a machine that should be running goes quiet or still, and ask what happened on the night shift.

See how it works

Ghost hunting cameras

Record every camera and microphone on an investigation, get flagged when it hears whispering, footsteps or a knock, and let an AI assistant find the moments worth a second look.

See how it works

Retail footfall

Count shoppers in and out, see where they stop and how long they queue, and set your takings against the footfall for a conversion rate.

See how it works

Occupancy and capacity

A live count of how many are inside across every door, a capacity limit, a screen for the entrance and an alert when it is nearly full.

See how it works

See all solutions