Scylla Introduces Event Raptor: An In-House Visual Language Model Built on Scylla NET

AUSTIN, TX – January 21, 2026 – Scylla Technologies Inc., a global leader in artificial intelligence powered physical threat detection and video analytics, today announced the release of Event Raptor, a proprietary Visual Language Model (VLM) developed entirely in-house and built on Scylla NET, the company’s foundational computer vision architecture.

Event Raptor is now available in open beta. Full integration into Scylla AI’s core platform and Alarm Hub is planned for the second quarter of 2026, where the model will support forensic video search, real-time alarm description, and natural-language threat recognition.

“Designed specifically for physical security and surveillance environments, Event Raptor represents a move away from cloud-dependent, general-purpose foundation models toward enterprise-grade visual intelligence that operates entirely under organizational control”, said Albert Stepanyan, Scylla’s Co-founder and CEO.

The model runs on-premises and in air-gapped environments, with no reliance on third-party APIs or external cloud infrastructure, ensuring that video data and security intelligence remain within the organization’s network.

Event Raptor is built directly on Scylla NET, Scylla AI’s proprietary computer vision architecture optimized using genetic algorithms. This approach enables high-performance visual understanding while maintaining a lightweight footprint suitable for edge deployment. The architecture supports real-time inference with sub-100 millisecond latency per frame and is designed to run on NVIDIA Jetson platforms, including Orin Nano and Orin AGX, as well as on-premises servers and isolated systems.

Unlike general-purpose VLMs trained on internet-scale image data, Event Raptor was trained on Scylla AI’s proprietary real-world surveillance datasets. These datasets span diverse environments, camera types, lighting conditions, and threat scenarios commonly encountered in security operations. As a result, the model delivers reliable performance in low-light conditions, crowded scenes, occlusions, and dynamic environments, while reducing the risk of hallucinations in high-stake security contexts.

Event Raptor recognizes and describes more than 1,000+ object, behavior, and situational classes, enabling detailed scene understanding beyond basic object detection. The model can generate operationally relevant descriptions of security events, providing security teams with immediate context and actionable intelligence.

When integrated into Scylla AI’s platform, Event Raptor will enable natural-language forensic video search across archived footage, automatic enrichment of real-time alarms with contextual descriptions, and the creation of custom threat recognition rules using plain language. These capabilities are designed to improve situational awareness, reduce operator workload, and accelerate response times without requiring coding or cloud connectivity.

Event Raptor is designed for deployment at the edge, with a compact footprint measured in gigabytes rather than terabytes. All video analysis and decision-making occur locally, minimizing network exposure and supporting strict cybersecurity and compliance requirements. This architecture is particularly suited for government, defense, critical infrastructure, healthcare, financial institutions, and other privacy-sensitive environments where external data transmission is restricted or prohibited.

By delivering a fully proprietary, edge-deployable Visual Language Model, Scylla AI provides organizations with a new option for advanced video analytics—one that combines modern AI capabilities with full control over data, infrastructure, and long-term model evolution.

Organizations interested in early access, custom tuning, or enterprise deployment can contact Scylla Technologies Inc. representatives directly or via our contact form.