Get up and running
fast.

This guide walks you through installing LookoutNVR, connecting your first camera, and enabling AI detection. The whole process typically takes 10-15 minutes.

Windows (10, 11, Server 2019+, 64-bit): Download the installer from the download page and run it. It asks where recordings should go, installs the recorder as a Windows service, and installs the LookoutNVR viewer alongside it.

The viewer opens when setup finishes. The first time, it asks you to create an administrator account - a username and password of your choosing. There is no default password to look up and none to change later.

To watch the same recorder from another Windows PC, install just the viewer on that machine and point it at the recorder's LAN address:

Two ways to install the recorder. Windows uses winget. Linux uses Docker. The app you watch on is separate from the machine that records.

Windows recorder. Package LookoutNVR.Server. It runs as a Windows service on Windows 10, Windows 11, and Windows Server 2019, 64-bit. Those packages are not in the winget catalog yet, so until Microsoft lists them, use the installer on the download page. It is the same program.

Windows viewer. Package LookoutNVR.Desktop, on any Windows PC that should watch a recorder. The server installer also puts a viewer on the recorder itself. Same wait for the winget listing.

Linux recorder. Docker image ghcr.io/lookoutnvr/lookoutnvr-server. The CPU tag is edge, and it is the smaller image. Use it when the machine has no NVIDIA card. The GPU tag is gpu, the larger image, and only on a machine with an NVIDIA card, driver 580 or newer, and the NVIDIA Container Toolkit. Do not point a machine with no NVIDIA card at the gpu tag. The image is public, so a pull does not need a login. The desktop viewer does not run on Linux. Point the Windows viewer at https://that-machine:5000.

Mac, and a Linux desktop. The Mac desk client can sign in and show live, review, objects, and alerts. The native Mac window is not built yet, so there is no Mac download. A Linux desktop app would be that same client in a Linux window, and that is not built either. Watch a Linux recorder from the Windows viewer.

How many machines. Add each camera's width times height times the frames you will decode. That total, plus detection, is the CPU. A solid desktop is often the whole recorder. A second building is a second recorder. If one building does not fit on one computer, split the cameras. Work the number on the sizing page. Examples are on the use cases page.

Lookout writes ordinary files into a folder. It does not speak NFS, SMB, or S3. If the operating system has already mounted a disk or a share, that folder is a normal path and the recorder can use it.

New footage. Prefer an NVMe or a SATA SSD in the computer. Cameras write all day, and Review reads the same files. On Windows, use a second drive such as D:\Recordings. Do not use the system drive root or a folder sitting directly under it. On Linux, mount a real disk at the recordings folder. Do not leave footage on the container's own small disk.

Older days. A second folder can take finished days after the copy is checked. Review still opens them. That is the place for a spinning disk, a larger SSD, or a network share. On Linux the share is an NFS or SMB mount. On Windows it is another drive letter. Do not point the live recordings folder at a share. A stalled share must not be the only place a new file can land.

How long it keeps. With no day limit and no size cap, the disk fills and then the oldest finished day is deleted. A day limit or a size cap deletes sooner. Setup asks for the one recordings folder. The second folder is a setting the recorder already honors. It is not a button on that first screen. A cloud copy of every camera, all day, is not in this release.

Go to Settings → Cameras → Add Camera. You have two options:

• ONVIF Auto-Discover: Click "Scan Network" and LookoutNVR finds cameras automatically on your subnet.
• Manual RTSP: Paste your camera's RTSP URL directly. See camera compatibility for URL formats by brand.

After adding, select a recording mode: Continuous, Motion-triggered, or AI-event-only. The camera appears in your live view immediately.

Motion is recorded first. Detection names what moved. Detection never decides whether the recording exists. It runs while something in the picture is moving, not on every frame of every camera, so a windy tree line costs more CPU than an empty hallway.

On Windows, detection uses DirectML when a graphics card is present and the CPU when it is not. On Linux, the edge image uses the CPU. The gpu image uses CUDA when the card is visible to the container, and falls back to the CPU when it is not. Turn detection on per camera. A card with less than 4 GB of memory can decode and still cannot run the accurate detector. No detector has been timed on a graphics card yet, so plan that CPU on the sizing page.

Your cameras are recording, AI is running, and events are being logged. From here you can set up motion zones, configure push notifications, or set up storage retention policies.