Home » Argu Products » Vision Agents / Chat with security cameras (Flagship Product

Vision Agents / Chat with security cameras (Flagship Product

Argu is cctv analytics you build with a sentence: describe the situation that matters, and an agent watches for it around the clock.

The difference is architecture, not a bigger detector list. The platform combines large language models with visual encoders, so it reads a scene closer to the way a person does: who is present, what they are doing, and whether that matches the rule set for that camera. A competitor's list of forty detection types is a ceiling. Here, the operator writes the forty-first one. Terminal operators run these agents across gate lanes and quay perimeters as part of a wider port security programme.

Key facts:

A scenario written in plain English becomes a live agent in about 30 seconds, without a developer writing a line of code – the same claim made above for this platform.
The platform connects to existing RTSP or ONVIF Profile S/T/G streams, IP cameras and NVRs, rather than requiring new detector-specific hardware.
Camera feeds run at standard 1080p–4K resolution and 15–30 fps, over either H.264 or H.265 compression, the two codecs most gate and perimeter cameras already use.
Most cameras on a network draw power over IEEE 802.3af (15.4W), 802.3at (30W), or 802.3bt PoE, so a new agent typically does not require new camera-side wiring.
Argu closed a two million dollar pre-seed round led by Speedrun, the early-stage fund from Andreessen Horowitz, with the Port of Miami and Mekorot investing as customers who run the platform themselves.
Terminal operators at ports fold these agents into a wider critical-infrastructure posture that can include NDAA Section 889 hardware restrictions at federally connected sites.

The same agent architecture is covered in more depth on Argu's vision agents page, and terminal operators run it as part of the wider port security programme referenced above. Government and critical-infrastructure teams layer it into homeland security surveillance, and teams comparing platforms can see how this differs from a fixed-detector product on Argu's best AI video analytics companies page.

Axis	Fixed detector catalogue	Argu vision-agent approach
What raises the alert	A match against one of roughly forty pre-trained detector classes	A scenario described in a sentence by the operator, turned into a live agent in about 30 seconds
What a human must watch	Every trigger from the fixed detector list, regardless of relevance to the site	Only the specific scenario written for that camera or zone
Integration path	A new engineering ticket or model retraining per use case	Existing RTSP or ONVIF Profile S/T/G streams, IP cameras and NVRs already installed
Applicable standards context	The detector list is fixed regardless of site-specific policy	An operator-authored rule can be scoped to a site's own policy, relevant when footage is reviewed under frameworks such as CJIS for law-enforcement users
Camera / feed required	Compatible only with cameras the vendor's detector was trained against	Standard 1080p–4K feeds at 15–30 fps over H.264 or H.265, on PoE-powered (IEEE 802.3af/at/bt) cameras already on the network
What gets retained	Detection logs tied to fixed categories only	Alerts plus the underlying live and archived footage, searchable by the same natural-language query

What is Vision Agents / Chat with security cameras (Flagship Product

Video analytics is software that reads live or recorded camera footage and turns it into events a security team can act on. Most video analytics and cctv analytics tools ship as a fixed menu of detectors: person, vehicle, loitering, and little else. This platform works differently. Describe the situation in plain English, the same way you would brief a new guard, and the sentence becomes a live agent in about 30 seconds. One operator built an agent that flags any employee who takes longer than 30 seconds to approach a customer after they walk in, without a developer writing a line of code. For a plain-language primer on how AI cameras and classic detectors differ, see our guide.

Key Takeaways

These are the core principles that define how Argu approaches every deployment. Our solutions are built to work on your existing infrastructure, require no engineering overhead, and deliver measurable results from day one. Every feature is designed with one goal: giving your team the intelligence to act faster, with more confidence, and without adding complexity.

See it in Action

Type a rule in plain English

It becomes a live agent in about 30 seconds

One dashboard covers every connected camera

How It Works

Describe the scenario

Write the rule in plain English, the way you would brief a new employee: what should happen, and what should never happen, in front of a specific camera. That description can be scoped as tightly as a specific zone or camera, or as broadly as every connected feed on the site.

The sentence becomes an agent

A large language model paired with a visual encoder turns the sentence into a live monitoring agent, without writing code or retraining a detection model. No new detector-specific hardware or camera swap is required; the agent reads the existing RTSP or ONVIF Profile S/T/G stream.

The agent watches every connected feed

Once live, the agent applies the same rule across every camera it is assigned to, correlating events across entrances, aisles or perimeters instead of watching one screen at a time. Correlation runs across feeds at standard 1080p-4K resolution and 15-30 fps, regardless of individual camera codec, H.264 or H.265.

An alert arrives in words

When the described situation occurs, an alert explains what happened in plain language, so a duty manager can act on it without reviewing an hour of footage first. That alert can reference the specific camera, zone, and timestamp where the described scenario was matched.

Product Advantages

Frequently Asked Questions

What is cctv analytics on this platform?

It is cctv analytics configured with a sentence instead of a settings menu. Describe the scenario that matters and the platform turns it into a live agent that watches connected cameras for that specific situation, in place of a fixed list of pre trained detectors.

Do we need to replace our cameras?

No. The platform connects to RTSP streams, IP cameras and NVRs that are already installed, and runs as a software layer on top of that network. Most sites keep their existing hardware and add this layer on top of it. Feeds are read at standard 1080p-4K resolution and 15-30 fps over H.264 or H.265, and most cameras already draw power over IEEE 802.3af, 802.3at, or 802.3bt PoE, so no new camera-side wiring is typically required.

How long does it take to build a new agent?

Writing a scenario in plain English and turning it into a live agent takes about 30 seconds. Connecting the camera feeds themselves depends on the site and is a separate, one time step.

How is this different from ordinary video analytics software?

Most video analytics software ships with a fixed catalogue of detection types to choose from. This system is built around large language models and visual encoders, so an operator can describe a scenario that is not on anyone's pre built list and get an agent for it the same day.

Who is already using this?

Argu's technology runs across four continents with governments, intelligence communities and critical infrastructure operators. The Port of Miami is a public customer reference, and is also one of the company's investors. Argu closed a two million dollar pre-seed round led by Speedrun, the early-stage fund from Andreessen Horowitz.

What video codecs and camera protocols does the platform support?

It connects over RTSP and ONVIF Profile S/T/G, reading standard 1080p-4K feeds at 15-30 fps encoded in either H.264 or H.265, the two codecs most IP camera and NVR systems already use.

Can a written scenario be scoped to a specific site's policy?

Yes. Because the operator writes the rule instead of picking from a fixed detector list, an agent can be scoped to a site's own policy, which matters for regulated environments such as law-enforcement footage reviewed under CJIS.