One infrastructure, infinite experts.
Our platform reads the scene the way a human would; it reasons and decides. Every expert — from OHS to retail — is this core in a different configuration.
RECPLATFORM · EXPERTSThe platform at work
The eye opens; the platform starts reading different worlds. A safety violation, a no-scan at the register, a defective PCB — one intelligence, different expert eyes.
Classic computer vision recognizes objects and stops: 'a person, a forklift.' Primarch understands the scene: who is doing what, what isn't normal, what happens next.
Holistic scene reading
The stream is split into real-time windows; a prefilter hands the moments that matter to the platform's deep analysis layer. The output is not a label — it is a reading at the level of action: 'An operator is reaching into the mixer, guard open, still energized.'
- Time windows, not single frames — the before and after of an action are read together
- Parallel real-time processing across cameras
- Findings in natural language: what happened, who did it, why it matters
Configurable expert architecture
An expert is defined by configuration: domain documentation, regulations, thresholds. The core never changes — which is why a new industry isn't months of model training, but a package that ships in days.
- Domain knowledge base + rule templates + industry thresholds
- Targeted fine-tuning when needed — never training from scratch
- Unlimited experts on the same core
Industrial integration & deployment
It connects to the RTSP cameras you already own; no new hardware. On-premise, footage never leaves your facility — compliant anonymization is part of the design.
- On-premise or cloud — data sovereignty stays with you
- Compliant anonymization and access control
- Dashboards, alert channels and a natural-language assistant
What every expert finds ready
Real-time multi-camera
Dozens of streams watched in parallel; the windows that deserve attention are prioritized automatically.
Natural-language findings
Every detection is a human-readable interpretation of the scene — an explanation, not a label.
Evidence clips
A time-stamped video segment attached to every finding, so everyone sees the same thing.
Queryable event memory
Expert findings are written into a facility memory; the past can be queried in plain language.
Alerts & reporting
Instant notifications, periodic digests and role-based dashboards — signal, not noise.
API & integration
Findings flow into your existing systems: ERP, messaging, work orders and BI tools.
Does it work with our existing cameras?
Yes. Primarch connects to your existing RTSP-capable IP cameras. No new cameras or special hardware required.
Where is our data processed?
Your choice: fully on-premise or in the cloud. With on-premise deployment, footage never leaves your facility, and compliant anonymization is applied.
How fast can a new use case go live?
Because experts are defined by configuration, typically within days — no months-long data collection or model training.
How is this different from classic computer vision?
Classic systems detect objects and stop. Primarch reads the scene holistically: it interprets actions and context, writes its findings in natural language and stores them in a queryable memory. Expert modules run on this core with domain knowledge on top.
Ready to make your cameras think?
A 15-minute live demo — with a scenario tailored to your industry.