The Physical AI platform

Cortex.
The Physical AI platform.

The on-device intelligence behind every Sevyn8 deployment. It senses, understands, decides, and acts in under 100ms — fully offline — then learns across the whole fleet.

One platform across retail, cold chain, logistics, and the home. Seven capabilities run as one system on the device, on any silicon.

Same event · different decision
Detection
Person at door — 3:17 AM
Context
Unknown face · Unusual hour
Decision
⚠ Alert homeowner
How Cortex works

Sense. Understand. Decide. Act. Learn.

Every event runs the same loop on the device — from raw signal to decision in under 100ms, then back to the whole fleet as learning.

Raw feed
HAL
AI Engine
Decisioning
Action
Raw feed
4K video, thermal, IoT, and sensor streams enter on the device.
01
Sense

Every camera, thermal, and sensor feed is ingested on the device — vision, heat, presence, sound, location — fused into one situation.

02
Understand

Quantised models read the scene in context: who, what, where, and whether it matters — against learned baselines, in under 100ms.

03
Decide

Industry vocabulary turns detections into decisions — shrinkage, breach, proximity, fall — the call a human would make, made instantly.

04
Act

Alert staff, open a counter, slow a forklift, log compliance — the right action fires on-device, with no cloud round-trip.

05
Learn

Every outcome feeds system-wide learning and returns to every device over the air. The fleet gets smarter with each deployment.

Seven capabilities · one system

Cortex runs all seven capabilities
as one system, on the device.

Sensing, understanding, deciding, and learning — together, on the endpoint. That's what Cortex does, and it's what makes physical AI actually deliver.

01 · HAL

A hardware abstraction layer normalises across boards — so new silicon is a supply decision, not an engineering rewrite.

01HALone API, any board
02SoC-agnosticARM · x86
03Universal data ingestionany sensor, fused
04Quantised AI on endpoints< 100ms · offline
05Industry vocabularyspeaks your domain
06Fleet learningevery device, OTA
07Privacy & authby architecture
The eighth — system-wide learning
Beyond the seven: every deployment teaches every other over the air. The fleet compounds — the part a competitor can't copy with faster hardware.
Enterprise & governance

Easy governance, by architecture.

The platform a security and compliance team can sign off on — MCP-native, cloud-agnostic, multi-tenant, and auditable end to end.

MCP-native

Three Model Context Protocol servers — Cortex Core, Edge, Admin/Ops — expose the platform to agents and external systems without custom glue.

Cloud-agnostic

Deploys on GCP, AWS, or Azure — same platform, same guarantees. Managed SQL, warehouse, object store, and event bus on whichever cloud you standardise on.

Bi-temporal data model

PostgreSQL 15+ with bi-temporal columns on every table. Every fact is queryable at any point in time — a complete, auditable history.

RBAC & multi-tenant

Row-level security scopes every tenant; role-based access controls every action. Multi-tenant and white-label ready across the fleet.

Edge inference

INT8-quantised vision and sensor-fusion models. Under 100ms, under 2W, fully offline. HAL abstracts ARM and x86 SoC families.

Compliance

DPDP + GDPR by architecture. STQC-aligned. Consent-based data flows. Privacy by design, not by policy.

See Cortex run on your hardware.

New devices or existing infrastructure — same platform, same AI. Measurable results in 60 days.