Conntour secures $7M for AI security video search engine

The video surveillance sector is navigating a complex moment, marked by intense public debate over privacy, safety, and the appropriate use of monitoring technology. High-profile cases involving law enforcement access to private camera networks have placed the industry under scrutiny. Yet, the market continues to expand, propelled by significant advancements in vision-language models that are creating powerful new tools for security and monitoring.

For Conntour, a startup building an AI-powered video search engine, this ethical landscape is central to its business strategy. Co-founder and CEO Matan Goldner states the company is highly selective about its clientele, a position he acknowledges might seem unusual for a young firm. This selectivity is possible, he explains, because Conntour has already secured several major government and publicly-listed customers, including Singapore’s Central Narcotics Bureau. “Having such significant customers allows us to stay in control,” Goldner said. “We select who uses our platform based on our judgment of what is both moral and legal, ensuring we understand and approve of the intended use cases.”

This early market traction has attracted investor confidence. The company recently closed a $7 million seed funding round led by General Catalyst, with participation from Y Combinator, SV Angel, and Liquid 2 Ventures. Goldner noted the process was remarkably swift, with the round finalized within 72 hours after an intense series of meetings.

The core of Conntour’s platform is its application of AI to transform how security teams interact with video feeds. It functions as a real-time search engine for surveillance footage, allowing personnel to use natural language queries to locate specific objects, people, or events across live or recorded video. The system can also autonomously monitor feeds for predefined threats and generate automatic alerts. This approach moves beyond legacy systems that rely on rigid, pre-programmed parameters. Instead, it leverages natural language processing and computer vision to offer greater flexibility. A user could ask, “Show me every instance of a person in a red jacket entering the rear stairwell,” and the platform would scan relevant footage to provide video results.

A key differentiator, according to Goldner, is scalability. The platform is engineered to manage thousands of camera feeds efficiently. He claims the system can process up to 50 feeds using a single consumer-grade GPU, like an Nvidia RTX 4090. This efficiency is achieved through a multi-model architecture where the algorithm dynamically selects the optimal combination of models and logic systems for each query, minimizing computational demands. The software is designed for flexible deployment, operating fully on-premises, in the cloud, or in a hybrid model, and can integrate with existing security infrastructure or function as a standalone solution.

The company also addresses a perennial challenge in surveillance, video quality limitations. Poor lighting, low resolution, or obstructed lenses can render footage nearly useless. Conntour’s system accounts for this by assigning a confidence score to its search results. When source video quality is subpar, the platform will still return potential matches but will flag them with a low confidence rating to inform the user’s assessment.

Looking ahead, Goldner identifies a central technical hurdle, balancing expansive capability with operational efficiency. “We face a contradiction,” he explained. “We aim to provide the full, unrestricted flexibility of a large language model, allowing users to ask anything. Simultaneously, we must maintain extreme efficiency to process thousands of feeds without prohibitive resource costs. Solving this contradiction is the foremost technical challenge in our field and our primary focus.”