ARSA Rules: Natural-Language Video Analytics Rules Software
Write what you want your cameras to watch for in plain language, and ARSA Rules checks it on your own GPU servers. A fast detector filters each stream, an on-premise vision-language model checks the rule, and every alert arrives with an evidence clip. Nothing leaves your network.
Self-hosted on your GPU servers · Evidence clip with every alert · Validated on your own footage
"Alert me when anyone stays in the loading bay for more than five minutes." The alert carries the clip that triggered it.
What A Rule Can Check
Zone presence and dwell time
Steps done in the right order
A person handling an object
Free-text rules, validated with you
Evidence clip with every alert
Checked on your clips before go-live
The Four Rule Types
Zone Presence And Dwell
Mark an area on the camera view and say what should or should not happen there: nobody in the loading bay after hours, no vehicle parked in the fire lane, nobody waiting at the gate for more than ten minutes.
Step Sequence
Check that a procedure happens in order: a worker signs the log, puts on gloves, then opens the cabinet. A step skipped or done out of order raises an alert.
Person-Object Interaction
Watch for a person and a thing together: someone lifting a crate off a restricted rack, touching a control panel, or leaving a bag behind at a counter.
Free-Text Rules
When a request fits none of the three types, you describe it in a sentence and we turn it into a rule. Free-text rules are validated for each customer on that customer's own footage. We make no accuracy promise for them: you see how the rule behaves on your clips, and you decide whether it goes live.
How A Rule Is Checked
Rules works in two stages. A fast detector watches each stream for the people, vehicles and objects a rule cares about, and only the moments that match go to a vision-language model on your own servers, which reads the scene and decides whether the rule is broken. When it is, you get an alert with the clip that triggered it.
- Detector filters the stream
- Vision-language model checks the rule
- Alert with evidence clip
On-Premise Video Rules For Your Own Servers
Rules software suits organisations that already run GPU servers and want the video, the models and the alerts to stay inside their own network. No servers? Run Rules on the AI Box Server 12 or 25 instead. Rules sits alongside the ready-made analytics in ARSA AI Video Analytics Software, for the requests a fixed detector cannot cover.
Runs on your own GPU servers
Installed on hardware you own and manage, in your data centre or server room.
Video never leaves your network
The detector, the vision-language model and the evidence clips all stay on site.
Alerts where your team already works
Each alert goes to a webhook, the dashboard or email, with its clip attached.
IMPLEMENTATION
How A Rule Goes Live
- Your camera
- Your GPU server
- Alert with evidence clip
Describe
Tell us, in your own words, what each camera should watch for. Our engineers turn every request into a rule and pick the rule type that fits it.
Validate
Each rule is run on clips from your own cameras before go-live. You review what it flagged and what it let pass, and the rule is adjusted until you are satisfied.
Run and route
The validated rule runs on your servers. Every alert carries its evidence clip and goes to your webhook, the dashboard or the email addresses you choose.
LICENSING
Rules Software Pricing
| Subscription | from $95 per camera per month, billed annually |
| Perpetual licence | $2,850 per camera, perpetual |
Rules software is licensed per camera, either as a subscription billed annually or as a perpetual licence, and the licence covers every rule that camera runs. Validating each rule on your own clips is part of getting started. Current list prices for every ARSA product are on the pricing page; for a site with many cameras, talk to us about the rules you need first.
MINIMUM
Technical Requirements
Rules runs on your own GPU servers, because both the detector and the vision-language model run on site. Your servers need at least an 8-core CPU, 32 GB of RAM, a 256 GB SSD for the system and a GPU with 24 GB of VRAM, the same minimum as ARSA Sovereign software; we recommend a GPU with 48 GB of VRAM or more. In many cases the cameras you already have are enough; the placement and resolution each rule needs are set out in our camera requirements.
| Where it runs | Your own GPU servers, inside your network |
| Video input | Your existing IP camera streams |
| Alert outputs | Webhook, dashboard and email, each with an evidence clip |
| Rule validation | On clips from your own cameras, before go-live |
| Data location | Video, clips and alerts stay on your servers |
Send Rule Alerts Where Your Team Works
An alert is only useful if the right person sees it in time to act. Each one names the rule, the camera and the time, and carries the clip that triggered it.
Webhook
A JSON event posted to an endpoint you choose, so alerts can open tickets or reach the incident tools you already run.
Dashboard
A list of every alert by rule and camera, where your team plays back each evidence clip.
Alerts sent to the people on duty for each rule, with the clip attached or linked.
FAQ
Questions About Video Analytics Rules
What can a rule check?
Four kinds of things: whether someone or something is inside a zone and for how long, whether the steps of a procedure happen in the right order, whether a person interacts with a particular object, and free-text rules for requests that fit none of those. Rules raises alerts for people to act on. Stopping machinery stays with your guarding, interlocks and emergency stops.
How is a free-text rule validated?
On your own footage. You send clips that show the situation the rule should catch and clips where it should stay quiet, our engineers run the rule on them, and you review every result. We adjust the rule until you are satisfied, and it goes live only when you agree. We publish no accuracy figure for free-text rules, because how well a rule works depends on your cameras and your scenes.
What hardware do we need?
Your own GPU servers, meeting the published minimum GPU specification, plus the IP cameras you already have. If you have no servers to spare, the same rules run on the AI Box Server 12 or Server 25. Camera placement guidance is on the system requirements page.
How do alerts reach our systems?
By webhook, as a JSON event posted to an endpoint you choose, on the Rules dashboard, or by email to the people on duty. Every alert names the rule, the camera and the time, and carries the evidence clip.
What happens to our video?
It stays on your servers. The detector, the vision-language model and the evidence clips all run and live inside your network, and no footage is sent to ARSA or to any cloud service.
How is Rules licensed?
Per camera, as a subscription billed annually or as a perpetual licence. One camera licence covers every rule that camera runs. Current list prices are on the pricing page.
How do we start?
Send us the rules you have in mind and a few clips from the cameras involved, through the free AI assessment or by applying to become a design partner. We tell you which rule type fits each request, run it on your clips, and show you the results before anything is installed.
Become A Rules Design Partner
We are opening Rules to a small group of design partners. A design partner gets rules built and validated on its own clips, early access to new rule types, and a direct line to the engineers who build them. Tell us what your cameras should watch for and we will come back to you.
Not ready to apply? Send a few clips through the free AI assessment or check your cameras against the system requirements.