Live in production

Live in four weeks. Measured against your baseline.

Physical AI,
in production.

We’re building the operating system for the physical world. It starts on the cameras you already own. Live today at one of India’s largest airports.

Scroll

The status quo

1,000,000,000

Cameras installed on earth

Watching nothing.

Footage rots on disks. Sensors fire into the void. Operations still run on someone walking the floor. The physical world is fully instrumented and completely blind. It does not need more hardware. It needs a brain.

<1%
Ever watched
≈0%
Acted on
24/7
Still recording

Everyone else is building the future of physical AI.

We’re already running it.

Most labs rehearse in simulation. This one runs in production: 600+ cameras, real operations, every decision logged. The future is not a demo.

The operating loop

Not a dashboard. A loop.

LEARN01PerceiveCameras + sensors02DecideAgents apply policy03ActWorkflows + actuation04AccountLogged + provable

Every step on infrastructure you own. Audit-logged end to end.

What it is made of

Five capabilities. One story.

Vision reads the floor. Sensors fill the gaps. Agents decide and do the work. Dashboards keep people in command. One loop, five capabilities.

Perceive

Computer vision

Every camera becomes a sensor that understands what it sees.

Perceive

IoT + sensors

UWB, BLE, LoRaWAN, RFID, access control, facility systems. Whatever the site speaks. Signal beyond sight.

Decide + act

Agents

Apply your policy, run the workflow, file the paperwork. With a camera in the loop or without one: the same runtime runs voice and document agents, with the same audit trail.

Oversee

Dashboards

One live view of operations. People stay in command.

Act · next

Robots

The same runtime will drive hands in the field. Next, not yet.

Live now

Running today at one of India’s largest airports.

CAM 01· Concourse
LIVE · 00:42:17
DESK
14
In queue
CAM 02· Retail floor
LIVE · 00:42:17
GAP
1
Stock-out flagged
CAM 03· Arrival ramp
LIVE · 00:42:17
280/h
Flow rate
CAM 04· Parking · ANPR
LIVE · 00:42:17
AP 09 CX 4412
B-12
Plate matched · bay
600+
Cameras on one edge fleet
4
Systems in production
4wk
From access to live
0
Third-party AI APIs

Operations platform

Hundreds of camera feeds fused with live flight and ground-handling data into one operations view. Queues, baggage flow, service times.

Live

Revenue leakage control

POS cross-referenced with camera in real time across dozens of retail stores. Discrepancies flagged with the clip attached.

Live · Expanding

Ramp congestion

Threshold and dwell alerts on the arrival ramp. One edge box covers roughly 40 cameras.

Live

Parking enforcement

IoT detects the vehicle, the PTZ camera swivels, ANPR reads the plate, the challan generates end to end. Approved by the operator.

Live

Customer name, architecture detail, and measured outcomes are shared under NDA. Reference calls arranged for qualified buyers.

Where it runs

One runtime. Every venue.

Every venue gets all three rows. SEE is what the cameras and sensors read. DO is what the agents handle. TRUST is how it stays governed in your world.

Airports

Live
See

Security and check-in queues · Baggage flow · Apron and ramp congestion · Retail floors · Parking occupancy

Do

Re-roster counters before the SLA breaks · Reconcile POS against camera · Auto-challan unauthorized parking · Flag the slow turnaround with evidence attached

Trust

KPI-gated pilot. Baseline captured first, results measured against it.

Hospitals

See

ED and OPD queues · Bed and OR turnover · Patient flow across wards · Falls · PPE and infection control

Do

Draft pre-auths, claims, and codes · Call housekeeping the minute a bed empties · Page rapid response with the clip · Prep nursing handover

Trust

No facial recognition. DPDPA processor posture. Pilots run in one department under your medical board.

Retail

See

Shelf stock-outs · Planogram and share of shelf · Footfall and checkout queues · Billing leakage

Do

Raise the replenishment order before the shelf runs dry · Open a counter before the queue forms · Reconcile POS against camera · File the dispute with the clip attached

Trust

Priced per store. Pilot in 50 stores, measured against matched control stores.

Cinemas

See

Concession and entry queues · Lobby density · Auditorium turnaround between shows · Parking

Do

Trigger staffing before the interval rush · Dispatch housekeeping when the credits roll · Reconcile F&B billing · Dispatch incidents

Trust

Runs on your existing cameras and POS. Priced per property, payback tracked per quarter.

Venues + campuses

See

Crowd density at gates and attractions · Flow across zones · Parking and shuttles · Perimeter

Do

Meter the gate before crush risk builds · Dispatch security and medics with location · Stage shuttles to demand · Reconcile ticketing and F&B across outlets

Trust

Crowd-safety thresholds set with your team. Rolled out zone by zone on your existing cameras.

Manufacturing

See

PPE and zone safety · Line stoppages · Visual quality checks · Forklift and pedestrian conflict · Yard and gate activity

Do

Open the work order in SAP PM with the clip · Hold the lot and flag QA · Sequence trucks at the gate · Draft the shift handover

Trust

Zone-level reporting, not worker tracking. Read-only on OT. Never part of the certified safety chain.

Cities

See

Traffic and junctions · Violations · Crowd events · Encroachment · Waterlogging · Streetlight outages

Do

Violation to challan with court-ready evidence · Dispatch and track incidents to closure · Meter crowds at events · Verify waste pickup and flag missed routes

Trust

Deployed as OEM through your systems integrator. Evidence chain built for court. Human review before issuance.

No rip and replace

Your cameras. Your systems.

Deployments start on the camera estate you already own and write into the systems your teams already use. New hardware only where it earns its place.

We read from
ONVIF / RTSP camerasYour existing VMSSensors: UWB · BLE · LoRaWAN · RFIDMQTT · Modbus · OPC-UAAccess controlFlight and ops feeds
We write into
HIS / EMR (HL7, FHIR)POS and billingSAP PM / CMMSDMS / SFAeChallan + ICCC APIsEmail, WhatsApp, ticketing

Read-only by default. Any write-back is gated by your policy and logged.

Security + compliance

Compliance is infrastructure, not a checkbox.

Your security team has a questionnaire. This section answers the first page of it before the NDA. Reference architecture, audit posture, and assessment detail follow under NDA.

Data residency
Customer-controlled. On-premise inference for sensitive workloads; for cloud or hybrid deployments, data and model live inside your cloud account. No third-party AI APIs. Inference never leaves your walls.
Data isolation
Single-tenant by design. Your footage, events, and tuned models stay yours. Nothing is pooled across customers, and nothing from your deployment improves anyone else's.
Identity handling
For vision workloads: appearance-based tracking (clothing, posture), session-local only. No facial recognition. No biometric templates stored.
Integration
MCP-compatible tool registry. Standard protocols where they exist (HL7 / FHIR for healthcare, OData / REST for enterprise systems). Read-only by default. Any write-back gated by explicit human sign-off.
Outbound alerts
Alerts you route outward (email, WhatsApp, ticketing) carry only what your policy allows. Evidence clips attach only where you permit them, and every send is logged.
Updates
Signed releases, delivered into your environment. Staged rollout with rollback. Model updates are treated like firmware: versioned, pinned, and reversible.
Model hosting
Self-hosted on infrastructure you control. On-prem GPU and edge for sovereign deployments, your cloud (AWS, Azure, GCP) or hybrid where that fits. Edge nodes for perception where camera latency matters. A central server (local or cloud, your choice) for inference, application APIs, and the audit database.
Compliance
Audit log for every model decision. Mapped to ISO 27001 controls. Exports in DPDPA (India), HIPAA (US), and SOC 2 formats.
Co-existence
Sits alongside your existing AI and cloud vendors. Open APIs. No data lock-in.
Support
On-site engineering presence during deployment. India-timezone support and SLAs after go-live.

We build the whole stack.

Silicon to decision. The cameras, the edge boxes, the models, the agents, the audit trail. Designed together, deployed as one. Inference never leaves your walls, and your data and models are never pooled with anyone else’s.

CamerasSensors · IoTEdgeOn-prem computeModelsDetect · trackAgentsPolicy · workflowsAuditEvery decisionSILICON TO DECISION · INFERENCE NEVER LEAVES YOUR WALLS

Zero-touch provisioning

A new edge box enrolls itself. No engineer on a ladder with a laptop.

Signed updates, rollback

Models and software ship as signed releases. Staged rollout, instant rollback.

Remote health

Every box, camera, and model reports in. Failures page us before they page you.

One console, any fleet

Designed to run one site or three hundred from the same control plane.

The master plan

01

Run the loop at airport scale.

Live
02

Run the same runtime in hospitals, retail, cinemas, factories, cities.

In motion
03

Run the edge fleet under every deployment.

By construction
04

Become the layer the physical world runs on.

Next

Steps 1 and 3 are not the plan. They are the head start.

How it starts

Three steps. The first is free.

01Free

Walkthrough

Bring one workflow and a camera layout. We map the loop onto your floor and tell you exactly what we would measure. A working session, not a sales call.

024 weeks

Pilot, measured

One site or one department. KPIs agreed up front, baseline captured before anything switches on. First system live in four weeks.

03Same runtime

Scale

Next workflow, next site. Each deployment starts with the integrations, audit trail, and lessons already in place.

Book a walkthrough

Systems integrator? In government and large programs we deploy as the OEM behind your bid. Your customer, your paper, deal registration. Partner with us.

Tell us what you’re running.

A short note about your site and the systems you run is enough for a first reply. The walkthrough is free.

Book a walkthroughcontact@apexflo.ai

Co-founded by Madan Lolugu (built AdeptChips, acquired by Quest Global in 2022) and George Veneel Dogga (IIT Guwahati, IIM Ahmedabad, ex-Microsoft, ex-Flipkart).

Hyderabad, India · Building the operating system for the physical world