অক্ষি OKKHI

On-premise vision AI

000Loading

Loading OKKHI

Public Infrastructure — OKKHI deployment

অক্ষি  ·  On-Premise Vision AI · Dhaka

Pioneering AI.
Transforming how business sees.

OKKHI builds the intelligence layer for modern enterprise — AI that watches every camera, inspects every unit, and understands every floor, entirely inside your walls. Digital transformation that starts with what you already own.

100% ON-PREMISEEDGE-FIRST RUNTIMEBANGLA + ENGLISH OCR
Proven builders

Engineered by a team with real production deployment experience — built right the first time, not learning on your floor

On-premise

Zero footage leaves the building — inference, storage, and alerting all run inside your network

Bilingual

Bangla and English text read natively for plates, labels, and container codes

One runtime

35 solutions share one hardened platform, so your second use case costs a fraction of the first

02 · Our work

Built for the field.
Not the lab.

Every tile below is a production-ready system, engineered by a team that's shipped this kind of AI before — open any of them to see the workflow, the demo, and the deployment spec.

340+

Cameras engineered to run on a single server

24/7

Continuous operation, lights-out reliable

<1s

From event to verified alert on-premise

100%

Of footage stays inside your walls

Know in seconds — not hours later.

The moment something goes wrong — a reject, a fall, a breach — the right person gets the alert with the snapshot attached. No scrubbing footage after the damage is done.

Your footage never leaves your walls.

Every frame is processed and stored on hardware inside your premises. Data-residency and compliance requirements are met by design, not by promise.

Your cameras. Our intelligence. Nothing replaced.

OKKHI runs on the CCTV you already own. One server watches hundreds of feeds, and a typical deployment is live in weeks — not quarters.

03 · The difference

Intelligence at the edge.
Privacy by architecture.

Most vision vendors stream your cameras to their cloud. We do the opposite: decode, inference, decision, and storage all run on a server inside your premises. The only thing that leaves is the alert you choose to send.

STAGE 01

Connect

We attach to your existing IP cameras over RTSP. Each camera gets its own resilient worker that reconnects automatically.

STAGE 02

Understand

A micro-batched inference engine shares each GPU across hundreds of streams; temporal voting confirms events before alerting.

STAGE 03

Act

Verified events flow to your dashboard, control room, or ERP. Every event is journaled locally first — nothing is ever lost.

See the full architecture →

04 · Why OKKHI

Global platforms weren't built for here.
We were.

Most vision platforms sold in Bangladesh were designed for someone else's networks, languages, and regulations. The differences show up in production — usually at the worst possible moment.

Typical cloud analyticsOKKHI edge-first
Your footageStreams to vendor cloud, often abroadProcessed on a server inside your walls — nothing leaves
Internet outageDetection stops; events are lostDetection unaffected; events journaled and delivered when the link returns
Bangla text & platesLatin-only OCR misreads or skipsDual-script reading, designed for it from the first line of code
False alarmsPer-frame triggers page your team all nightTemporal voting plus on-premise VLM verification before any alert
Tampered cameraGoes unnoticed until footage is neededCovered, defocused, or moved cameras flagged in seconds
Server billA server per few dozen streamsHundreds of streams per GPU server — a fraction of the hardware

05 · Industries

Powering the industries
that power Bangladesh.

Our engineering team has shipped production vision systems across FMCG, banking, ports, and heavy industry. We build for the realities of this market: unstable networks, mixed camera fleets, Bangla-language environments, and zero tolerance for data leaving the country.

IND/01

Banking & ATM

Branch counters, vaults, and ATM booths — tampering, vandalism, and suspicious-entry analytics, processed on-premise.

IND/02

Pharmaceuticals

Blister, fill, cap, and label inspection for export-grade GMP compliance — audit trails for every unit that leaves the line.

IND/03

FMCG Manufacturing

High-speed bottling and packaging lines: counting, fill verification, and label QC without slowing throughput.

IND/04

RMG & Heavy Industry

PPE compliance, fire and smoke detection, and restricted-zone monitoring across factory floors and yards.

IND/05

Ports & Logistics

Container code reading, gate automation with dual-script plate recognition, and yard surveillance at depot scale.

IND/06

Public Infrastructure

Road, traffic, and facility analytics for agencies — engineered for sovereign, on-premise deployment from day one.

06 · Questions

The questions every
operations head asks first.

Do we need to replace our existing cameras?

No. OKKHI is designed to run on the CCTV you already own over standard RTSP. We assess your current fleet during the pilot and tell you honestly if any specific camera placement needs adjusting — most don't.

Where does our video data actually go?

Nowhere. Decoding, inference, decision-making, and storage all happen on a server inside your premises. The only thing that leaves your network is an alert you explicitly configure — an email, an SMS, or a webhook to your own system.

How long until we see results?

A pilot runs about two weeks: roughly a week to install and calibrate against your cameras and conditions, then a week of measured operation on your real footage. You get a written report with the measured detection performance at the end.

What happens if the internet goes down?

The system keeps working. Because nothing depends on a cloud service, detection and journaling continue uninterrupted, and any queued alerts are delivered once connectivity returns.

How do you avoid flooding our team with false alarms?

Two layers. Temporal voting requires a detection to hold across consecutive frames before it counts, and for high-stakes alerts an on-premise vision-language model verifies the event before a human is paged. The system fails closed, never open.

What does a deployment cost?

It depends on camera count, the solutions you need, and whether you use existing hardware. We scope it during the pilot and give you a fixed written proposal — no per-camera cloud subscription, no surprise egress fees.

Still have a question we haven't answered?

Let's Talk
Free 2-week pilot · you keep the report Let's Talk

Proof beats promises.
See it on your own cameras.

We run a two-week pilot on your footage, with your incidents, measured against your numbers. If the results don't convince you, you keep the full report and owe nothing.

Book a live demo

For teams deploying vision AI

Field notes from real deployments.

Once a month: what we learned shipping AI monitoring on factory floors and branch networks in Bangladesh — what worked, what broke, what we'd do differently. No product pitches.

No spam. Unsubscribe any time.