Restricted-area entry
Entry into a polygon you draw — server room, cash office, tank farm, substation, bonded store.
Access control tells you a door opened. It does not tell you who walked through it.
Security intelligence for cameras you already own
Blepo connects to the cameras already on your wall, proves they are actually working, turns what they see into incidents a person can act on, and keeps the record of what happened next.
No rip and replace. No proprietary cameras. One piece of software that speaks ONVIF and RTSP, reads without ever writing to your recorder, and tells you plainly when it cannot prove something rather than colouring it green.
Blepo connects over RTSP and ONVIF, which most of your estate already speaks. Recorder-level checks need a vendor adapter, and what any given adapter can actually prove depends on the model, the firmware and the access you grant it. Those four brands are candidates under test, not blanket guarantees — we publish a tested compatibility matrix per model rather than a wall of logos, and a camera we cannot prove something about is reported as UNSUPPORTED.
NO RIP-AND-REPLACE · NO NEW CAMERAS · NO RECORDER SWAP
Blepo checks every device on the site, every hour and signs the proof that it did.

THE PREMISE
Nigerian estates, warehouses and branches have spent a decade buying cameras. Very few of them can tell you, this morning, which of those cameras produced usable footage last night. That is the gap Blepo closes first.
Every AI security platform sells you detection and quietly assumes the footage is there. Nobody checks. So before a single detection runs, Blepo checks each camera independently — stream, image quality, recorder state where the recorder exposes it, and clock drift — and reports each check as verified, failed, unknown, unsupported or stale. Never as a single green dot.
Every camera polled on a schedule you configure. A stream that is unreachable, unusable or still serving the same frame is reported as failed, not quietly skipped.
Where a recorder exposes it, we confirm it is genuinely writing this channel — not merely powered on and answering pings. Where it does not, we say so.
On recorders whose adapter supports a retention test: how many days of footage are actually retrievable right now, per camera, against the policy you believe you have.
Detects a shifted view, a defocused lens, an obstructed frame or an IR failure against a reference frame.
Signed heartbeats from every site. Silence is treated as an alarm, not as an absence of news.
Drifted timestamps make footage useless as evidence. We measure the drift against a trustworthy reference and report it — correcting it is a change on your NTP, not something Blepo reaches in and does.
Verify runs on the cameras you already own, on hardware already at the site. No new recorders, no rip-and-replace, no capital approval and it is the reason every layer above it can be trusted.
Two states force every uncertainty into a lie. A camera whose recorder cannot be queried is not healthy — it is unverifiable, and you deserve to be told which.
A camera can be online, in focus, perfectly framed, and writing nothing to disk. Blepo reports stream health and recording state as two independent checks, and reports recording as verified only where a supported recorder returns affirmative evidence. Everywhere else it reads RECORDING UNKNOWN — because that is what we actually know.
Recording verified means a recorder-specific playback or retention test returned evidence inside a defined window. It is not a guarantee about future footage, and we do not present it as one.
A rule you draw on a camera view behaves the same way on day one and on day four hundred. A behavioural model does not, until it has seen your site. Those two things are sold as one thing everywhere else in this category.

Geometry, schedule and object class. You draw them on your own camera view and they do exactly what you drew, which is why they can ship first.
Entry into a polygon you draw — server room, cash office, tank farm, substation, bonded store.
Access control tells you a door opened. It does not tell you who walked through it.
A line crossed, optionally in one direction only. In but not out, or out but not in.
Direction is the difference between a delivery and a theft.
A person or a vehicle on site outside the schedule you set, per camera, per day of week.
Most losses happen when the site is empty and nobody is watching the screen.
Presence inside a marked zone for longer than a threshold you choose.
Reconnaissance looks exactly like loitering, and precedes most planned incidents.
Camera covered, frozen frames, or a major change to what the camera is pointed at.
Reported as an observation with evidence, never as an accusation of sabotage.
Every rule is drawn on your own camera view, on your own schedule, with a preview mode and a version history. Rules disable themselves automatically when the input quality is too poor to trust — a detection from a defocused camera is worse than no detection at all.
Related detections are correlated into a single incident with a severity, an explanation and whatever evidence the integration permits. Every underlying signal is kept — correlation groups the noise, it never discards it.
Every transition records who did it and when. Machine-generated candidate signals are shown distinctly from human-confirmed incidents, and marking one a false alert is captured as feedback rather than swallowed.
Blepo reads. It does not delete, overwrite or alter an original recording on your NVR, and the connector has no API for doing so.
Every item says whether Blepo is holding a copy or merely holding a pointer into footage your recorder owns. You always know which one you are relying on.
Event time, source camera, retrieval time, rule and model version, and an integrity hash where the integration supports one.
Tenant-scoped storage, short-lived links, configurable retention — and an audit trail of who opened what, which exists for your protection as much as ours.
Recipient groups, escalation chains, quiet hours and cooldowns — and delivery failures shown as failures. If WhatsApp drops the message, the incident is still open, still assigned, and still waiting for somebody.
Not because it is slow — because nobody has measured it on your cameras, your uplink and your recorder yet. We instrument event-to-incident, incident-to-queue and queue-to-provider timings from day one, and we will publish medians and 95th percentiles once pilot deployments have produced them. Any vendor quoting you a sub-second figure before seeing your site is quoting you a lab.
Every check Blepo runs is signed on your site, before it touches our servers and each batch is cryptographically linked to the one before it. Change a reading after the fact and every link after it breaks. That is the difference between a status page and evidence you can hand to an insurer, an auditor or a court.
The signing key never leaves your site. Even we cannot forge a reading that was never taken which is the whole point of asking us.
When the reading was taken and when it reached us are recorded separately. The gap between them tells you about your uplink, not just your camera.
The monthly report cites the exact stretch of the record it was built from, so the person receiving it can verify it rather than take your word.
Most security upgrades start with a quotation for hardware. This one starts with the estate you already paid for.
A lightweight agent runs on hardware already at the site and finds your cameras over ONVIF, or takes their stream addresses directly. No new cameras, no re-cabling, no recorder swap.
Every camera is checked continuously — stream, image quality, recorder state where the recorder supports it, clock. Nothing is analysed on a camera Blepo cannot first prove is working, and a check we cannot complete is reported as unknown rather than passed.
Mark perimeters, doorways, restricted areas and tripwires directly on the camera view, and set the hours each rule applies. Everything is normalised, so a resolution change does not invalidate your work.
Related signals correlate into one incident, the permitted evidence is attached, and it routes to the person on duty — in-app, on an approved WhatsApp template, or into your own systems by signed webhook. One tap marks it a false alert, and that feedback is measured rather than swallowed.
Nothing here is a separate product or a separate price. What changes is which rule gets drawn on day one — and that is a conversation, not a dropdown.
Yards are large, staffed thin at night, and full of things worth taking.
FIRST RULETripwire across the loading gate, inbound only, outside shift hours
02When an auditor asks for last Tuesday, the answer cannot be a shrug.
FIRST RULERecording verification on every branch camera, every day, with a monthly report
03Shrinkage is a measurement problem long before it is a camera problem.
FIRST RULEAfter-hours presence, plus merchandise-removal patterns in controlled pilot
04One guard, forty cameras, and a screen nobody has looked at since Tuesday.
FIRST RULERestricted-area polygons on plant rooms, dwell thresholds on perimeter roads
05A camera aimed at a wall passes every uptime check ever written.
FIRST RULEIntegrity checks first — tamper, obstruction, clock drift, agent silence
06The risk is in the corridors guests never see.
FIRST RULEBack-of-house zones and service entrances, scheduled around trading hours
Bring your existing footage. We'll show you what Blepo finds in it.