Aurora Flow analyses sequences of video rather than single frames, recognising actions and behaviours as they unfold. We built it to run fully on-premise and air-gapped across entire camera estates in near real time, and it's already live in enterprise deployments.

Today we're launching Aurora Flow, a frontier video understanding model purpose-built for live, enterprise-scale surveillance. Where most commercially available video AI analyses one frame at a time, Aurora Flow can read up to 60 seconds of continuous motion, so it recognises not just what is visible in a scene but what is actually happening. It's already deployed in live environments, runs fully air-gapped, and works across multiple cameras in near real time.

What is Aurora Flow?

Aurora Flow is a generative vision-language model that analyses video sequences rather than single frames. We built it for on-premise, air-gapped deployment with no cloud dependency, and it's native to Sentry, our video intelligence platform, working inside the detection and alert workflows our customers already run. That combination, sequence-level understanding delivered securely at scale, makes it one of the most advanced AI models commercially available for surveillance.

‍

‍

Why analysing motion matters

Traditional video analytics classifies still images pulled from a feed, and some events genuinely can be identified that way. Many cannot. Fighting, theft, tampering and falls unfold over time, and a single frame rarely proves what is going on. The result is familiar to anyone who runs a busy control room: too many alerts, too little context, and too much footage passed to manual review.

We built Aurora Flow to add the missing dimension. By recognising movement patterns and reading behavioural sequences as they unfold, it separates an incident from routine activity before an operator's time is spent.

How it works inside Sentry

Everything in Sentry starts with a trigger, a signal that a section of video needs a closer look. When one fires, for example when our self-learning engine flags a genuinely unforeseen event, Sentry applies layers of analysis to the footage. Aurora Flow is now the most advanced layer we can apply. It reads the motion and describes what's happening in plain language: "the scene shows an incident where someone appears to require medical attention", or "the scene shows routine activity requiring no action". Only one of those needs a person. Operators set their own rules for which readings become alerts, so control rooms see fewer, better ones.

‍

‍

From Aurora to Aurora Flow

Aurora Flow is an evolution of Aurora, the vision-language model that has sat inside Sentry's live alerting decision for the past 18 months. Where existing AI models see a snapshot, Aurora Flow sees a story.

Aurora

Reads single frames: objects, people and scenes.

Best for:

• Perimeter intrusion
• PPE compliance
• Static hazards and unsafe conditions

Aurora Flow

Reads sequences: actions, behaviours and events over time.

Best for:

• Behaviour and safety monitoring
• Incidents that a still image cannot confirm
• Complex event investigation

What it unlocks

Behaviours that once needed human eyes to catch can now be identified more accurately as they happen. Validated use cases so far include:

• Shoplifting and other theft detection across retail, hospitality and industrial settings
• Flashmob and malicious-loitering identification for retail and public safety
• ATM and equipment tampering in retail banking
• Unsafe-behaviour warnings in workplaces and public settings
• Enhanced left-object and abandoned-item detection for transport and public spaces
• Enhanced fight, fall and aggressive-behaviour detection, relevant to every sector

"We believe Aurora Flow is a frontier AI model in surveillance and a step change in what video intelligence can deliver, moving beyond detection and into genuine understanding of behaviours and actions in complex real-world environments. Things like fighting, climbing and theft have typically required human eyes to detect them accurately. Now we can help operators focus on what needs their urgent attention by putting AI into the alert decision."

Callum Wilson, Founder and Co-CEO

Frequently asked questions

What is Aurora Flow?

Aurora Flow is Eluviant's generative video understanding model. It analyses sequences of video, not single frames, to recognise actions and behaviours as they unfold, runs fully on-premise and air-gapped, and is native to Sentry, our video intelligence platform.

How is Aurora Flow different from traditional video analytics?

Traditional video analytics classifies still images taken from a feed, which struggles with events that unfold over time. Aurora Flow analyses up to 60 seconds of continuous motion, so behaviours like fighting, tampering and theft can be identified more accurately as they happen.

Does Aurora Flow need the cloud?

No. Aurora Flow deploys fully offline, on-premise and air-gapped, with no cloud dependency. That's how we meet data sovereignty and compliance requirements in secure and sensitive environments.

How does Aurora Flow relate to Aurora?

Aurora Flow is an evolution of Aurora, the single-frame vision-language model that has been part of Sentry's live alerting decision for 18 months. Aurora reads frames (objects and scenes); Aurora Flow reads sequences (actions and behaviours over time). They work together inside Sentry.

Is Aurora Flow available now?

Yes. We launched Aurora Flow on 14 July 2026 and it's already deployed in live enterprise environments. Talk to us about running it on your cameras.