Asilla Launches Full-Scale Deployment of VLM-Integrated AI Security System 'AI Security asilla' with Real-Time Video Analytics

Asilla Inc. (Headquarters: Machida City, Tokyo; CEO: Tsuyoshi Onoe; hereinafter 'Asilla'), which envisions a 'safe and comfortable world through the power of technology,' has built the technical foundation for the full-scale deployment of its Vision Language Model (VLM)-integrated AI security system, 'AI Security asilla,' using the NVIDIA Metropolis platform, starting in June 2026.

This foundation integrates VLM-based situational awareness into a large-scale streaming analytics platform that processes numerous camera feeds in real-time, all on the same hardware. This enables a comprehensive video analytics solution on a single platform, covering everything from 'detection' to 'semantic understanding.'

Development Background

In the field of video analytics, demand is growing not only for handling increased processing loads and infrastructure costs due to a rising number of cameras, but also for complex situational understanding that traditional detection models alone cannot capture.

VLM, an advanced AI technology gaining attention for such situational understanding, requires significantly more computational resources than conventional detection models due to its architecture that integrates visual and linguistic processing. Achieving this requires simultaneously optimizing large-scale parallel processing and inference efficiency to run VLM without increasing resources.

Asilla has solved this challenge by leveraging the various SDKs that make up the NVIDIA Metropolis platform. Preliminary deployment began in spring 2026, with full-scale rollout to large domestic facilities starting in June.

Key Features of the Platform

- Large-scale parallel streaming processing: Utilizes NVIDIA DeepStream SDK to process multiple camera streams simultaneously in real-time. - AI inference optimization: Maximizes throughput and hardware utilization efficiency by optimizing inference for multiple AI models using NVIDIA Triton Inference Server and NVIDIA TensorRT. - VLM integration on the same hardware: Leverages the resource headroom created by inference optimization to run VLM on the same hardware as detection models, providing both anomaly detection and semantic understanding of video content in an integrated manner. - Comprehensive video analytics: Offers large-scale parallel streaming processing, including VLM, as a single product, reducing deployment and operational costs.

Technical Architecture

The platform is built on NVIDIA Metropolis, using NVIDIA DeepStream SDK for large-scale parallel streaming pipelines, and efficiently executes inference for each AI model (behavior recognition models and VLM) using NVIDIA Triton Inference Server and NVIDIA TensorRT. This enables a configuration where detection models and VLM run concurrently within the same hardware resource limits.

Application Scenarios

The AI security system 'AI Security asilla,' leveraging this platform, is designed for large-scale facilities operating numerous cameras, such as commercial complexes, office buildings, and transportation infrastructure like airports and railways. By combining real-time anomaly detection from detection models with situational understanding from VLM, it contributes to on-site situation awareness and initial response support.

Future Developments

Asilla will further leverage the NVIDIA Metropolis platform, using the integrated VLM as a core to expand its analytical capabilities beyond conventional detection toward deeper situational understanding—answering not just 'what' and 'where,' but also 'in what context' events occur. The company will also continue improving detection accuracy for anomalies and precursors, and optimizing models for specific operational environments.

Executive Comment

【Masahiro Wakisaka, CTO, Asilla Inc.】

'We can now deploy the expertise we've cultivated in behavioral recognition AI on a large scale and efficiently through the NVIDIA Metropolis platform. By integrating VLM on the same hardware, we can offer customers video analytics that understands situations—not just detects anomalies—at a realistic cost.

Running VLM in real-time across numerous camera feeds is no easy feat; this configuration is a significant step forward in overcoming that barrier. Using this platform as a foundation, we will continue to expand the possibilities of video analytics.'

Related Press Releases

Asilla, AI for Video Analytics, Begins Full-Scale Deployment of Proprietary VLM 'AsillaVision' in Major Domestic Facilities

https://jp.asilla.com/post/news-asillavision-20260624

Asilla Selected for GENIAC, a National Project by Japan's Ministry of Economy, Trade and Industry and NEDO

https://jp.asilla.com/post/news-geniac-20260604

Availability

Full-scale deployment of the 'AI Security asilla' system, built on this technical foundation, began in June 2026. For details on deployment, please contact the inquiry point below.

*Note: NVIDIA, NVIDIA Metropolis, DeepStream, TensorRT, and Triton are trademarks or registered trademarks of NVIDIA Corporation in the U.S. and other countries.

About 'AI Security asilla'

A system where AI analyzes existing security camera footage 24/7, instantly detecting abnormal behaviors such as violence, falls, and intrusions, as well as attention-requiring behaviors like wandering, congestion, and signs of illness. Amid a severe shortage of security personnel, it captures anomalies easily missed by human monitoring and immediately alerts security staff and managers. Since it leverages existing cameras, no new equipment investment is required, enabling high security levels even with limited personnel—a next-generation security solution.

Asilla Inc.

CEO: Tsuyoshi Onoe

Headquarters: 1-4-2 Nakamachi, Machida City, Tokyo

Business: Development and provision of various products and solutions based on behavioral recognition AI

Official Website: https://jp.asilla.com/

Media Contact

Asilla Inc.

PR Contact: Nakamura

Email: [email protected]