Commercial property security monitoring solutions, sorted by how much of your camera plant they ask you to throw away

Recording is solved. Monitoring is the actual gap.

Every commercial property of any meaningful scale already has cameras, a DVR or NVR, and 30 to 90 days of retention sitting on a hard drive somewhere. Recording is solved. The thing that is broken is monitoring, which is a separate problem with a separate buying motion. Monitoring is what answers the question of who is looking at the feed during the 14 minutes of an actual incident.

Most operators conflate the two. They look at a flagged incident report, see grainy footage, and conclude they need better cameras. The fix that maps to that conclusion is to replace cameras, which is the most expensive of the five categories below and the one most likely to be the wrong tool for the actual job. Better cameras with no monitoring layer still produces footage that nobody watches. Worse cameras with an alert layer that pages a human in 4 seconds catches the same incident.

The five categories that follow all attempt to close the monitoring gap. They differ in cost, capex shape, infrastructure disruption, and what they require the property staff to do differently. The buying decision is choosing the category that fits the property profile, not picking a winner.

The five categories, defined

The list below is not ranked by quality. It is ranked by how much existing infrastructure each category asks you to remove. Live monitoring removes nothing. Self-hosted analytics removes nothing. The categories in the middle each remove a different layer of the camera plant.

1. Live human monitoring

On-site guard, overnight watch, or remote virtual guard service watching the feeds. The oldest category and the one with the highest judgment per incident, because there is a human on the other end. Cost scales linearly in operator-hours. A 16 hour overnight watch on a 40 camera property runs $2,000 to $4,000 per month for virtual guards, or $3,000 to $5,000+ per month for an on-site overnight guard.

Best fit: properties where a single missed incident costs more than $50K (luxury retail, hospital lobby, biotech facility, executive site), and properties with a regulatory or insurance requirement that names a credentialed human in the response loop.

Worst fit: portfolios with more than 6 to 8 properties, because cost is linear in property count and one guard cannot watch a portfolio. Mid-market multifamily, mid-tier commercial real estate, and self-storage operators have spent two decades trying to make the math work here, and most have given up on the overnight shift specifically.

2. Smart camera replacement

Verkada, Rhombus, and a handful of others. Replace every camera with a unit that has on-device detection and a vendor cloud control plane. Cleanest end state. Highest capex. The line item that sells the deal is the camera. The line items that sink the budget are the rewiring, PoE+ switch upgrades, licensed installer time, conduit work, the parallel-running bridge period, and the multi-month uplift in property staff hours.

A $200 per camera price tag turns into $1,000 to $2,500 per camera fully loaded. For a 60 camera property that is the difference between a $12K project and a $90K project. Recurring is $30 to $40 per camera per month, which is another $1,800 to $2,400 monthly on top of the bill.

Best fit: brand new construction with no existing camera plant, properties under 8 cameras where capex stays small, and operators who explicitly want a single-vendor end-to-end stack and have the budget to commit.

Worst fit: any property with a working recorder and 20+ cameras already installed. The cost curve is structurally wrong and the project ends up dragging for two quarters before the new system is fully cut over.

3. NVR replacement with bundled AI

Eagle Eye, Avigilon, Hanwha, and similar. Keep the cameras. Replace the recorder with a smarter one that adds analytics on top. Middle-ground capex. The cameras stay because most are RTSP-compatible, but the new recorder costs $5K to $15K depending on channel count, plus a per-channel licensing fee. Recurring lands at $500 to $1,500 monthly depending on camera count.

The hidden cost is the cutover. Removing the existing recorder means losing 30 to 90 days of footage in transit unless the bridge period is carefully managed, and the camera credentials sometimes need to be re-entered one camera at a time. For analog cameras over coax (still common in older Class B and C multifamily), this category often does not work at all, because the new NVR is IP-only.

Best fit: properties with all-IP cameras already, an aging recorder that needs replacement anyway, and an operator who wants a single new vendor for both recording and analytics.

Worst fit: properties with mixed analog-and-IP plant, properties whose existing recorder is healthy, and operators who want to preserve optionality. NVR replacement creates a new vendor lock-in at the recording layer.

4. Edge AI sidecar (HDMI ingestion)

The category Cyrano sits in. Leave the cameras alone. Leave the recorder alone. Plug a small box into the recorder's HDMI output, the same one already feeding the leasing office wall monitor. The box reads the recorder's tile grid (up to 25 feeds for a Cyrano unit), runs detection, dedup, threat-tier classification, and the SMS-and-call dispatcher locally, and pushes alerts and dashboard events upstream over the property network.

The architectural choice that defines the category is the HDMI ingestion path. There is no per-camera RTSP handshake, no codec negotiation per camera, and no auth dance per camera, because the box is downstream of all of that. If the recorder can render a tile grid for the wall monitor, the model can detect on it. Mixed brands, mixed resolutions, mixed analog-over-coax with IP, all work, because they all converge at the recorder before the box ever sees them.

Pricing for Cyrano specifically is $450 one-time for the device and $200 per month per property after the first month. One box covers up to 25 feeds, so a 25 camera property and a 12 camera property cost the same. No camera replacement, no rewiring, no IT contractor.

Best fit: commercial multifamily, mid-market commercial real estate, construction site CCTV trailers, self-storage, and any property where the recorder is healthy and replacing the camera plant is overkill. Most properties between 8 and 50 cameras land here.

Worst fit: properties whose recorder is so old or so locked-down that it does not produce a usable HDMI tile grid, properties with more than 50 cameras (would need multiple boxes and the math starts shifting), and properties with regulatory requirements that demand a human in every response loop.

5. Self-hosted RTSP analytics

Frigate, ZoneMinder, and custom builds on top of an in-house GPU server. The cost shape is the inverse of every other category: hardware and software are nearly free, but the labor is significant. Plan for one engineer-week of setup per property, ongoing model retraining as cameras and angles change, and a custody-of-the-system question when that engineer leaves.

Best fit: tech-adjacent owner-operators who already have an engineer who understands cameras, single-property operators who treat the system as a project rather than a service, and security-aware operators who do not want any of their video metadata leaving the property.

Worst fit: any organization without an in-house technical owner. Self-hosted is a verb, not a noun. The system that runs at month one is rarely the system that runs at month twelve, and somebody has to be paying attention.

What the sidecar path actually does on the wire

The HDMI ingestion choice is the part of the sidecar category that nobody outside of operator circles understands well, so it is worth pulling apart. The diagram below shows the message flow during one incident, from the moment a person crosses an alert zone to the moment the on-call manager's phone rings. None of the LAN traffic crosses the property uplink. Only the alert metadata and dashboard event do.

Two things to notice. First, the cameras and the recorder talk to each other on the LAN, the recorder talks to the sidecar over HDMI on the same LAN, and the sidecar runs every step of detect, dedup, and classify on its own CPU and accelerator. The internet is only on the egress path, which means a property uplink hiccup at 2 a.m. does not silence the alert chain. Detection keeps running, the local outbox queues up, and the drain worker walks events forward in strict order when the WAN comes back.

Second, none of this requires per-camera configuration. The box reads what the recorder is already drawing. That is what makes the sidecar category honest about working with mixed legacy plant. The other four categories on this list either ignore the existing plant entirely (smart camera replacement) or have to authenticate per camera (NVR replacement, self-hosted RTSP).

Mapping a property profile to a category

The cleanest way to choose a category is to answer four questions in order. Skip any of them and you end up buying on aesthetics.

What “works with everything” should actually mean

Every commercial monitoring vendor uses the phrase “works with your existing system” somewhere in their pitch. The phrase is doing different work in different categories.

• In smart camera replacement, “works with” usually means “we can run alongside the old system during the bridge period.” Translation: rip-and-replace, slowly.
• In NVR replacement, “works with” means “we accept your IP cameras over RTSP if their firmware is recent enough.” Translation: keep the cameras, replace the recorder, hope the firmware works.
• In sidecar (HDMI ingestion), “works with” means “if the recorder can drive the leasing office wall monitor, we can read it.” Translation: actually agnostic to the camera plant beneath.
• In self-hosted RTSP, “works with” depends on what your engineer can configure. Translation: it works with whatever you make it work with.
• In live monitoring, “works with” means a human watching the existing wall monitor. Translation: as compatible as anything could be, at the price of one human per shift.

The sidecar phrasing is the one that holds up under the most legacy plants. That is the practical reason this category has gained ground in the last 24 months for mid-market commercial real estate and Class B/C multifamily, where the camera plant is rarely uniform and rarely young.

Three honest cases where the sidecar is wrong

The honest version of any taxonomy includes the cases against the author's preferred category. Three failure modes for the sidecar category specifically:

One. The recorder does not have a usable HDMI output. Some old DVRs only output VGA, some only feed proprietary client software, and some output an HDMI signal that is over-saturated or weirdly cropped. Pre-pilot you should physically confirm the wall-monitor signal is a clean grid. If it is not, the sidecar category is structurally wrong and the right answer is NVR replacement.

Two. The property has 60+ cameras and one recorder. One Cyrano unit covers up to 25 feeds. Past 50 cameras you are looking at three boxes per property, and the per-property math starts approaching what a smarter NVR would cost. The break-even is not always obvious; sometimes three sidecars still beats an NVR cutover, sometimes it does not. Run the numbers.

Three. The property has a regulatory or insurance requirement that names a credentialed human in every response loop. Healthcare facilities, certain financial services properties, and some hospitality categories fall here. The sidecar handles detection but the response side still needs a human in the chain, and the cleanest answer is a hybrid where the sidecar pages the human instead of replacing them.