AlphaAnivar NVR · free and open source · Apache-2.0

It watches. It listens.
It remembers.

A free, open-source alternative to Blue Iris. Anivar records your security cameras around the clock and understands what they see and hear — the faces you teach it, number plates, breaking glass — then answers your questions about it. It's an ordinary app for your computer, and all of it runs there.

Download View on GitHub

Windows, macOS (Apple Silicon) and Linux

Sam · person Front door14:32:07 ABC 1234 ABC 1234 silver car Driveway14:35:51 Glass breaking 0.91 · heard twice Back garden14:41:12 Ask Anivaron this computer Anyone at the front door last night? One visit, at 23:14 — the front door. searched 18:00 – 06:00 · 1 clip Front door 23:14 · 0:42 person Unknown seen once Name them →
Illustration of what Anivar does, not a screenshot.4 cameras · 1 computer · 0 cloud
  • A normal installerNo Docker, no server, no terminal. Download it, open it, add a camera.
  • No accountThere's nothing to sign up for, and no Anivar cloud to sign in to.
  • No telemetryNothing about you or your footage is reported anywhere.
  • Your GPU, or just the CPUDirectML, CUDA and TensorRT on Windows. CoreML on Mac. CPU everywhere.

What it does

Recording is the easy part.

Anivar also knows what it's looking at. Every model runs on your computer, and none is downloaded until you ask for it.

TUE 23:14:07 playback 0006 1218 24

Every second, on your own disk.

Each camera records continuously to files on your computer, within a disk budget and retention you set. Scrub to any moment — events are pinned on the timeline.

  • USB webcams, RTSP, MJPEG, and ONVIF discovery
  • Sub-second live view for IP cameras over WebRTC
  • Export a clip, or an incident report with its video
Sam3 of 3 frames agree Ana4 of 4 frames agree Unknown→ Who is this? named only after frames agree

Faces you teach it. No one else.

Enrol someone once and Anivar names them when they appear. A name needs several frames to agree before it's used, and people it doesn't know wait in Review for you.

  • Face recognition runs on your machine
  • Person re-identification links sightings across cameras
  • Remove a person and their biometrics are erased
ABC 1234 silver car · Driveway Seen 3 times A recurring vehicle — name it? Name

Plates read. Cars remembered.

Number plates are read on-device and matched to the names you give them. A car that keeps coming back is flagged, so you can decide who it is.

  • Plate reading plus body colour
  • Known plates shown by name
  • Vehicles get their own feed in Review
pitch ↑ time → TOP 3 Glass breaking0.91 Speech0.22 Wind0.08 Glass breaking · Back garden confirmed in 2 windows in a row · 14:41:12

It hears what the camera hears.

On cameras that carry sound, audio is classified against 521 everyday sounds — glass, alarms, barking, a raised voice. Most must be heard twice in a row before they alert, so one bang isn't an emergency.

  • YAMNet sound classification, on-device
  • Top three sounds and loudness saved with each event
  • Sounds get their own feed
person with a package 09:41 11:05 18:52 Front door · 16:20 matched by what each detected object looked like

Ask for what you remember.

Search by what was in the frame — “red shirt”, “person with a package” — or ask in plain words. The answer comes from your recordings first, with the clips attached. When there's nothing there, it says so instead of guessing.

  • Image search over every detected object
  • A language model that runs on your computer — or your own OpenAI, Anthropic, Gemini, Groq, xAI or LM Studio
  • Every answer says which hours it covers
BEHAVIOUR RULES Zone entryLoiteringClimbing a fence RunningPerson downCrowd Your Anivar bot14:52 loitering Loitering · Front gate Someone is lingering at the gate. Acknowledge View clip

Alerts only when it matters.

Draw zones and fence lines on each camera, then choose what earns an alert: entering a zone, loitering, climbing a fence, running, someone falling, a crowd. Alerts go to your own Telegram bot, with quiet hours and per-category mute.

  • Motion masks for trees, fans and screens
  • Browse people, vehicles and sounds from Telegram
  • Share a clip as a signed link that expires

How it works

Everything happens on this computer.

There is no Anivar server. The app on your desk is the whole system — recording, recognition, search and the assistant.

Your computer
  1. CamerasUSB, RTSP, MJPEG, ONVIF
  2. RecordEvery stream written to your disk
  3. DetectONNX Runtime on your GPU or CPU
  4. UnderstandFaces, plates, sounds, behaviour
  5. RememberA database and video files in your app folder
  6. AnswerSearch, and a language model on your machine

Only these ever leave the computer — and only if you switch them on:

TelegramoptionalAlerts and clips, sent to your own bot.
A cloud AIoptionalOnly if you add your own API key.
TailscaleoptionalViewing links over your own tailnet, with an expiry.

Compared

How it stacks up, honestly.

Anivar is young. Here's where it leads today, and where others are still ahead.

FeatureAnivarBlue IrisAgent DVRFrigate
How you install itA desktop installerWindows installerInstaller, script or DockerDocker (recommended)
Runs onWindows, macOS (Apple Silicon), LinuxWindowsWindows, macOS, Linux, Raspberry PiLinux; macOS through Docker; Windows not officially supported
Source codeOpen · Apache-2.0ClosedCore not publishedOpen · MIT
PriceFree$39.95–$99.95Free locally; remote access and apps from $7.95/moFree; optional Frigate+ models $50/yr
Faces and platesBuilt inBuilt inWith a paid planBuilt in
Recognises sounds521 classesNot listedWith a paid plan500+ classes
Search footage by descriptionBuilt inNot listedThrough outside assistantsBuilt in
Language modelIncluded, runs on your computer — or bring your own keyNot listedOutside assistants and vision models you connectBring your own (Ollama, llama.cpp or a cloud API)
Home Assistant / MQTTMQTT, with Home Assistant discoveryMQTTBothBoth
Phone appTelegram bot and share linksiOS ($9.99) and AndroidiOS and Android, with a subscriptionInstallable web app
MaturityAlphaEstablishedEstablishedEstablished, large community

Checked against each project's own website and documentation in September 2026. Something out of date? Tell us and we'll fix it.

Private by construction

Your footage never needs to leave.

No account, no cloud

There's nothing to sign up for and no Anivar server to send anything to.

Models when you ask

Nothing is downloaded until you choose it — the detector, face and plate models, even the language model.

Biometrics you control

Face data stays in your app folder. Remove a person and their biometrics are erased completely.

Download

Get Anivar.

Free, open source, and yours to keep. Windows is where it's developed and tested day to day; macOS and Linux builds come from the same code.

Windows

Windows 10 or 11 · 64-bit

Download

Anivar-windows-x64-setup.exe

Installs for you only — no administrator prompt. Uses any DirectX 12 GPU, and NVIDIA cards through CUDA.

macOS

macOS 11 or later · Apple Silicon

Download

Anivar-macos-arm64.dmg

Intel Macs are not supported.

Every file has a SHA-256 checksum published beside it — release notes and checksums. To update, open Settings → Check for Updates in the app.

These early builds aren't signed by a certificate authority, so Windows and macOS each ask once before opening them. What to click.

Built in the open

Help build the camera app you'd trust.

Anivar is Apache-2.0 and developed on GitHub. There's work at every size, from a one-line fix to wiring in a new model — and every change gets a review.

  • Rust
  • Tauri 2
  • React
  • TypeScript
  • ONNX Runtime
  • llama.cpp
  • SQLite
  • ffmpeg
Good first issuesSmall, scoped, with the files named.
DiscussionsQuestions, ideas, show your setup.
RoadmapWhat's next, and what's deliberately not.
Contributing guideBuild it, test it, send a pull request.