Ambient.ai announced the general availability of Ambient Pulsar, its advanced AI engine and the first reasoning Vision-Language Model (VLM) for agentic physical security. Built on an edge-optimized architecture, Pulsar represents a quantum leap forward, transforming physical security through proactive incident prevention. Trained on over one million hours of ethically sourced video, this purpose-built vision-language model processes over 500,000 hours daily, exceeding the performance of OpenAI GPT-5 and Google Gemini 2.5 Pro in relevant use cases at up to 50× higher efficiency, and enabling true agentic AI at enterprise scale.
Pulsar: Ambient.ai’s Next-Generation Vision Language Model
Ambient.ai has released Pulsar, its next-generation Vision Language Model (VLM), engineered to transform physical security. This model operates continuously at the edge, combining visual perception, semantic understanding, and autonomous reasoning. Pulsar is purpose-built for physical security, processing over 500,000 hours of video daily after being trained on more than one million hours of ethically sourced enterprise video. This edge-optimized architecture allows for parallel inference across numerous live video streams utilizing the latest NVIDIA AI infrastructure.
Pulsar delivers frontier-model reasoning performance exceeding OpenAI GPT-5 and Google Gemini 2.5 Pro in physical security applications, with up to 50x higher efficiency. Key breakthroughs include open-set detection – recognizing an infinite variety of behaviors, even those not specifically trained – and contextual intent recognition, understanding why something is a risk. This model powers new capabilities like Agentic Video Walls, Activity Notifications, and Semantic Search, establishing a new benchmark for intelligent security.
The launch of Pulsar reinforces Ambient.ai’s vision of Agentic Physical Security, where intelligent systems augment human operators. Early results demonstrate significant impact for enterprise customers, including up to 95% false-alarm reduction and 80% of alerts resolved in under one minute. By unifying monitoring, threat assessment, and investigations, Ambient.ai aims to create a proactive security system that understands, prevents, and responds autonomously.
Key Capabilities of Agentic Physical Security
Ambient.ai’s Pulsar is a next-generation Vision Language Model (VLM) engineered for agentic physical security. Unlike traditional deep-learning detectors, Pulsar performs continuous reasoning at the edge, reducing cloud costs and latency. This edge-optimized architecture, powered by NVIDIA infrastructure, allows parallel inference across numerous video streams. Key to its capabilities is open-set detection, enabling it to recognize a limitless variety of behaviors and threats, even previously unseen anomalies.
Pulsar unlocks several new platform capabilities, including Agentic Video Walls, Activity Notifications, and Semantic Search. Agentic Video Walls transform static security feeds into a dynamic system highlighting the most relevant activity. Activity Notifications deliver context-aware alerts based on custom events defined using natural language. Semantic Search allows operators to query video archives using simple questions, instantly retrieving key insights from hours of footage.
The Ambient.ai platform, powered by Pulsar, is demonstrating measurable impact for enterprise customers. These include reductions of up to 95% in false alarms and resolving 80% of alerts in under one minute. The system unifies monitoring, threat assessment, investigations, and response, creating a security system that not only detects risks but also proactively prevents incidents and orchestrates autonomous responses.
With Pulsar, Ambient.ai has built what every enterprise security leader has been waiting for: an AI that’s not just fast, but intelligent.
Cary Monbarren, Senior Director, Corporate Security at SentinelOne
Benefits of the Ambient.ai Platform
Ambient.ai’s new Pulsar is a reasoning Vision-Language Model (VLM) designed to transform physical security. Unlike traditional systems, Pulsar operates continuously at the edge, reducing cloud costs, latency, and bandwidth demands. It’s capable of open-set detection, meaning it recognizes a wide variety of behaviors—even previously unseen threats—and understands why something is risky through contextual intent recognition. This edge-optimized architecture leverages NVIDIA AI infrastructure for parallel inference across numerous video streams.
The Pulsar platform introduces several key capabilities, including Agentic Video Walls which dynamically highlight the most important camera feeds, and Activity Notifications customizable via natural language input. Semantic Search allows operators to query video archives using plain language, instantly retrieving relevant footage. Early previews include Agentic Investigations, which assembles incident timelines automatically, and Custom Threat Assessment, which adjusts alert severity based on specific scenes.
Ambient.ai customers are already seeing measurable results with the platform, including up to 95% reduction in false alarms and 80% of alerts resolved in under one minute. Pulsar processes over 500,000 hours of video daily and exceeds the performance of models like OpenAI GPT-5 and Google Gemini 2.5 Pro in physical security applications, offering up to 50x higher efficiency and millions saved in operational costs annually.
