Features

Everything below is in the current release. Every feature is also documented in the built-in Help.

Cameras

• Records the camera's RTSP stream directly. Adding a camera by a plain RTSP URL, for cameras without ONVIF, is coming soon.
• Automatic setup for ONVIF cameras: discovery (WS-Discovery) and capability probing (Profiles S/T/G/M), with optional daily re-probing.
• Main and Sub stream selection, per-stream enable and rename, H.264 and HEVC, with audio.
• Multi-lens cameras are split into one camera per lens.
• Vendor integrations matched automatically by make and model (for example, Dahua / Amcrest smart events).
• Groups for sites, buildings, and floors.

×

Recording

• One Windows-service recorder node per machine, with as many nodes as you need. Recording is ffmpeg -c copy, so no re-encoding.
• Recording modes: Continuous, Motion, Schedule, and Event, set globally or per camera, with pre/post-roll.
• Retention set globally, per node, and per camera, plus per-camera quotas and a disk watermark backstop.
• An optional archive volume (SMB share or USB drive) that receives aged-out footage instead of deleting it.
• Recording failover to a backup node, and a maintenance mode.
• Self-updating nodes: each build is approved once, and then every node installs it.
• Nodes keep recording through a server or database outage, even across a reboot.

×

Detection & events

• Server-side motion detection with polygon zones or a mask grid.
• Camera events over ONVIF, with user-defined event tag rules and timeline colors.
• Built-in AI object detection (YOLOX, D-FINE, or your own ONNX model) on NVIDIA, AMD, Intel, or CPU, or an external HTTP inference service.
• Live bounding boxes, object badges on live tiles, and a cropped snapshot for every detected object.
• Object tracking (ByteTrack), confidence scores, optional high-resolution snapshots, and a Slice mode that tiles wide or panoramic cameras so small objects are still detected.

×

AI hardware

AI detection is optional and set per camera. A node without it still records, and still uses server motion, camera events, and vendor integrations. The node picks a backend at startup:

Models are never bundled. The built-in YOLOX models are downloaded from the server on first use, and you can drop any .onnx model onto a node. 5–7 fps per camera is enough for accurate detection and tracking.

Tested hardware

LarisVMS has been run in real environments on:

• Windows: Windows 11, Windows Server 2022, and a Windows Server 2022 guest VM on Hyper-V 2022 with RTX 2070 GPU passthrough
• Recording storage: NVMe, Hyper-V virtual disks (on NVMe), a Windows Storage Pool passed through to a Hyper-V VM, and SMB file shares
• AMD Ryzen 7 5800X3D
• Intel Core i5-12600K, including object detection on its integrated GPU via DirectML
• NVIDIA GeForce RTX 2070: the main testing GPU and a dedicated recording node (DirectML, CUDA, and TensorRT FP32)
• NVIDIA GeForce RTX 4080 Super (DirectML, CUDA, and TensorRT FP16)

Viewing

• Saved camera-wall views with drag-and-drop layout, a fullscreen kiosk mode, and rotation.
• Adaptive live streaming that uses Sub streams for small tiles.
• Synchronized multi-camera playback on a zoomable timeline, with preview thumbnails and 1/32× to 32× speed. Switch any live tile to playback in place.
• Multi-camera exports, bookmarks, and a snapshots browser.
• Basic PTZ controls.
• Video relayed through the server, through media proxies, or sent directly from node to browser.
• A phone-friendly layout, installable as an app from Chrome on Android or Safari on iOS, with pinch-to-zoom on timelines and in fullscreen.
• Dark mode.

×

On your phone

The whole interface works in a phone browser, with the sidebar behind the ☰ button. To install it as an app, open it in Chrome on Android and choose Install app from the ⋮ menu, or in Safari on iOS choose Share → Add to Home Screen. It opens full screen, under your branding name. Installing needs a certificate the phone trusts; see Use it on your phone.

Administration

• Roles, a permissions matrix, and per-camera/group access control, with auto-expiring role assignments and PTZ priority between roles.
• Optional Microsoft Entra ID sign-in.
• Alert rules that watch a camera or node, sent by email (SMTP, Microsoft Graph, Gmail), webhook, ntfy, Pushover, Slack, and Teams.
• Audit log, system logs, scheduled database backups, and a health dashboard with per-camera fps, bitrate, and reconnects.
• API keys for automation, and a monitoring status endpoint.
• IP allow lists, an optional separate port for video traffic, and secrets encrypted at rest.
• Branding: app name, colors, font, and logo.
• Built-in Help with documentation for every feature.

×

Known limitations

• Windows only, for both the server and the recorder nodes.
• Cameras are currently added through ONVIF. Adding a camera by a plain RTSP URL is coming soon.
• Privacy-mask burn-in and camera-side motion zones can be configured, but are disabled pending fixes.
• Bounding boxes from the camera's own analytics aren't shown; only LarisVMS's own detection draws boxes.
• PTZ and Entra ID sign-in haven't been tested against real hardware or a real tenant.
• Active Directory integration is in the works.
• Some features listed may have placeholders (webhooks, Teams/Slack integration)
• Basic email functionality should work with SMTP. Graph API email has not been tested yet

Found something else? Please open an issue.