Qualcomm Insight Platform: Unified control, edge-first, AI-powered Video SaaS

What you should know:

• The Qualcomm Insight Platform embeds native AI directly into the video stack, transforming raw footage into a continuously queryable, real-time intelligence layer. This fundamentally changes how video is utilized for safety, security and operational efficiency.
• Our edge-first architecture ensures AI inference runs on-device or on local edge appliances, keeping raw video securely on-site. This design optimizes privacy and bandwidth by transmitting only metadata and specific events upstream.
• Merlin, our embedded generative AI agent, provides an intuitive, human-language interface for querying video streams. This capability allows the platform to deliver deeper insights and flexible deployment options, moving beyond traditional VMS and bolt-on analytics.

Native AI, not an add-on

Most incumbents have taken legacy video management solution (VMS) architectures and layered AI "add-ons" on top, often as separate services that are loosely coupled, difficult to scale and limited in capabilities. Qualcomm Insight is built around native AI from the ground up: AI models are embedded into cameras, edge boxes and the control plane, so understanding video is a first-class function, not an afterthought.​

At the core, Qualcomm Insight fuses high-performance vision models (convolutional neural networks and transformer-based detectors) with Vision Language Models (VLMs). This allows the platform to translate human language into structured queries over time, space, objects and behaviors in the video stream, without shipping all video to a central cloud. As video volumes continue to explode, this native AI architecture is crucial for precise, targeted queries instead of coarse, scan-all-footage workflows.​

Merlin: profile-aware, built-in GenAI assistant

Merlin, the embedded generative AI agent in Qualcomm Insight, is the primary interface for operators and systems. Unlike generic chat-style assistants or simple "AI search bars" added to legacy VMS, Merlin is tightly coupled with the video data model, policies and profiles defined in the platform.​

Traditional systems might support requests like "show me any person wearing a red shirt," relying on basic attribute filters layered on top of object detection. Merlin can support more sophisticated, profile-based queries such as "find me Joe wearing a red shirt between 3 p.m. and 5 p.m. across all entrances," combining identity or profile metadata, appearance attributes, time windows and camera topology. This level of specificity is essential for emerging security use cases where "any person" or "any vehicle" is not enough; operators need to reason about specific individuals, assets and behaviors in context.​

Merlin also:

• Generates incident summaries and reports directly from video and metadata, reducing manual report-writing overhead.​
• Supports profile-linked alerts, so notifications can trigger only when specified entities or behaviors match a given risk model, rather than flooding operators with generic motion or object alerts.​

Edge-first architecture for real-time, private inference

Qualcomm Insight is built around an edge-first architecture. Inference runs on-device or on local edge appliances, keeping raw video within the site and sending only metadata, events and selectively requested clips to upstream systems. This design reduces latency for real-time decision-making, preserves privacy, and dramatically improves bandwidth and storage efficiency.​

Qualcomm Technologies' heterogeneous compute consisting of CPU, GPU, NPU and ISP working in concert, allows tight integration of image signal processing, classical computer vision and deep learning on a single processor. This enables:​

• Real-time analytics at the camera or appliance without dedicated server racks.
• Multi-stream processing for high-density deployments.
• Efficient execution of both discriminative models (detection, tracking, classification) and generative models used by Merlin.​

Brownfield and greenfield deployment models

Qualcomm Insight supports both brownfield and greenfield deployments so customers can modernize at their own pace.​

• Brownfield: For sites with existing low-cost, non-AI cameras, the platform uses Edge AI boxes to terminate RTSP/ONVIF streams and run AI locally, preserving the existing camera and cabling footprint. Video is analyzed at the edge, with only metadata and event clips propagated to the Qualcomm Insight control plane, enabling a "no rip-and-replace" transition to a modern VSaaS architecture.​
• Greenfield: For new builds or full refreshes, the platform works with on-device AI cameras that run inference directly on Qualcomm Technologies vision processors. These cameras act as first-class, managed endpoints, coordinating with Merlin and the Qualcomm Insight backend for policies, profiles and analytics, while minimizing backhaul and edge-appliance footprint.​

With both models under a single management and analytics fabric, customers can mix legacy and new sites while maintaining common policies, query semantics and AI capabilities.​

Expanding camera flexibility with Augentix acquisition

Qualcomm Technologies, Inc. also completed the acquisition of Augentix to further strengthen the Qualcomm Insight hardware ecosystem. Augentix provides low-power ISP and multimedia SoCs optimized for smart cameras, enabling rich image processing and on-device AI. With this integration, Qualcomm Insight customers gain access to a broader portfolio of smart cameras from cost-optimized to feature-rich using silicon tuned for video analytics workloads.​

This wider choice allows system designers to tailor camera selection to each zone: simple low-power devices for low-risk areas, higher-performance AI cameras with advanced analytics and better low-light performance for critical locations, all under the same Qualcomm Insight control plane. The result is a more flexible, cost-efficient design without fragmenting software stacks or analytic capabilities.​

Multi-vertical and future-ready

Qualcomm Insight is architected as a horizontal platform with vertical specializations, not a set of isolated point solutions. Key examples include:​

• Standard security and surveillance for buildings and enterprises: weapon detection, loitering and tailgating detection, restricted zone entry, delivery personnel and package tracking, as well as automated detection and alerting based on configured profiles and policies.
• Smart cities and transportation: Cross-intersection tracking, incident detection and traffic flow analytics with edge processing at poles or roadside units and profile-aware searches during investigations.​
• Retail and logistics: Loss prevention, queue and dwell analytics, and zone-based monitoring, with the ability to search for specific individuals or assets based on role, profile or visual attributes.​
• Industrial and critical infrastructure: PPE compliance, restricted-zone breaches and anomaly detection in industrial lines, with video tightly coupled to OT systems and policies.​

Because AI is native to the platform, Qualcomm Technologies can continuously update models, add new behaviors and expand vertical templates through update pipelines without re-architecting the system. As video data grows and use cases become more complex, the Qualcomm Insight profile-aware Merlin agent and edge-first AI design provide the precision and scalability that bolt-on AI solutions struggle to match.​

We are actively working with a growing ecosystem of hardware and software partners to accelerate time-to-value for our customers, from pre-validated camera SKUs to out-of-the-box analytics bundles. Over the next several weeks, we'll introduce new capabilities and ecosystem collaborations that extend what Qualcomm Insight can do.