Self-hosted on Ubuntu
Your server, your disks, your GPU, your footage. No cloud account to record.
Ividiyo installs on Ubuntu 24.04 LTS as a set of systemd services, discovers your ONVIF cameras, and records them to local storage. The AI detection runs on a GPU in the same machine. Nothing is uploaded anywhere.
The phrase gets used loosely enough to be worth pinning down. In Ividiyo, every part of the system runs on hardware you own: the recording, the AI inference, the search index, the database, the web interface and the storage. There is no component that needs to reach the internet to do its job.
Pull the uplink and the site keeps recording, keeps detecting, keeps raising alarms and stays fully usable from inside the network. What you lose is remote viewing from outside and the ability to check a licence — and the licence check is designed to tolerate an outage rather than disable a working system. For sites where the fibre is the least reliable thing on the property, that distinction is the whole argument.
It also means the storage question is yours to answer. Recent footage lives on local SSDs for fast scrubbing, with optional NAS or archive pools behind them for long retention, and retention is set per camera rather than as one global number.
Ordinary, boring, well-understood components. Nothing here is exotic, which is the point.
Three services — management, recorder and the streaming server — started by systemd and restarted automatically if they fail. Logs go to journalctl where your existing monitoring can already read them.
Recording never re-encodes. Frames are copied from the camera's own stream into fragmented MP4 segments, so CPU use stays low and quality is bit-for-bit what the camera produced.
The segment index, events, zones, users and audit trail. A standard Postgres you can back up with the tools you already use.
Each camera is republished locally once and shared between recording and viewers, so twenty people watching a camera is still one connection to it.
Live walls, playback, forensic search, alarms and administration. Nothing installed on an operator's machine, no per-seat licence, and iOS and Android apps included.
Object detection and plate recognition on an NVIDIA GPU, a Hailo HAT or a Coral accelerator. Recording works fine without one and AI can be added later.
Three shapes that cover most of what people actually build.
Home and small sites, up to roughly 8–10 cameras. 8 GB of RAM, Raspberry Pi OS 64-bit, recordings on a USB 3.0 SSD. Add a Hailo AI HAT on a Pi 5, or a Coral USB accelerator on a Pi 4, for detection.
Up to a few hundred cameras. Ubuntu 24.04, 16 GB or more of RAM, SSDs for recent footage, and one mid-range NVIDIA card — which carries detection on 150+ cameras on its own.
Hundreds to thousands. Recording servers are added as the site grows, all under one management interface, with federated multi-site and automatic failover between recorders on Enterprise.
The installer is a web wizard rather than a configuration file. Download the package for your architecture, run the installer, and it takes you through creating the admin account, pointing at a storage path, and discovering cameras — then hands you the browser interface.
ONVIF discovery sweeps the network and finds what is there, pulling stream URLs, resolutions and capabilities from each camera, including whether it can report its own motion events. Cameras that are not discoverable can be added by RTSP URL. From there it is choosing which cameras record continuously, which record on motion, and how long each keeps footage.
Most sites are recording within an hour and spend the rest of the afternoon on detection zones and alarm rules, which is the part worth taking time over.
Worth knowing before you download rather than after.
Ubuntu 24.04 LTS, 64-bit, with Python 3.12 and PostgreSQL 16 — both of which the installer sets up. On ARM it is Raspberry Pi OS 64-bit on a Pi 4 or Pi 5.
Yes. Recording, AI detection, playback, search, alarms and the whole web interface run locally and keep working with the uplink unplugged. Internet is needed only for viewing the site from outside your network and for licence activation.
It depends on resolution, frame rate and retention, but a 4 MP camera recording continuously at a typical bitrate uses roughly 1–2 TB a month. Recording on motion rather than continuously cuts that substantially — on a measured lab site, raw motion covered about 43% of the day against 24-hour recording. Retention is set per camera, so the cameras that matter can keep footage far longer than the ones that do not.
Yes. Recording and playback do not use it, so a site can run on CPU for months and add a card when the AI features are wanted. The detection runtime is installed afterwards from the settings page rather than requiring a reinstall.
No. Video is written to your disks and never leaves them. Remote viewing streams from your own server to your own browser or phone over an end-to-end encrypted session, and the relay that helps the two find each other cannot read what passes through it.
The Free tier runs up to 10 cameras with live view, recording, playback and clip export — no licence key, no time limit, no card. Point it at the cameras you already own and see what it does before anything changes.
Need something to send on? Download the product brochure (PDF) — it carries this comparison and the pricing.