The surveillance technology sector currently faces intense scrutiny, largely due to privacy concerns. Recent controversies include the U.S. Immigration and Customs Enforcement’s access to Flock’s camera network for monitoring individuals, and criticism directed at home camera manufacturer Ring for developing features that would allow law enforcement to request neighborhood footage from homeowners. These events have fueled a significant public discourse surrounding safety, privacy, and accountability in surveillance practices.
Despite these controversies, the market for surveillance solutions remains robust. The continuous evolution of vision-language models is actively boosting companies that are developing innovative tools for organizations to monitor their premises effectively.
Matan Goldner, co-founder and CEO of the video surveillance startup Conntour, emphasizes the critical importance of ethics in this field. He states that his company is highly selective about its clientele. While this might seem counterintuitive for a startup just two years old, Goldner attributes this ability to Conntour’s existing portfolio of significant clients, including several large government entities and publicly traded companies, such as Singapore’s Central Narcotics Bureau.
“Having such substantial customers empowers us to be selective and maintain control. We actively manage who uses our technology and for what specific purposes, allowing us to choose partners we deem ethical and, of course, legal. We apply our full judgment, making client decisions based on a clear understanding of how they intend to utilize our system,” Goldner revealed in an exclusive interview with TechCrunch.
Conntour’s notable early success extends beyond its ability to choose clients. It has also attracted considerable investor attention, recently concluding a $7 million seed funding round with participation from General Catalyst, Y Combinator, SV Angel, and Liquid 2 Ventures.
Goldner mentioned that the funding round was finalized remarkably quickly, within 72 hours. “I scheduled approximately 90 meetings over about eight days, but within just three days – starting Monday and finishing by Wednesday afternoon – the round was complete,” he shared.
Conntour’s cautious approach to client selection appears justified, especially given the increasing sophistication of AI tools in this domain. The company’s proprietary video platform leverages AI models to enable security personnel to search camera feeds in real time using natural language, effectively functioning as a Google-like search engine for security footage. The system can also independently monitor for and detect threats based on predefined rules, automatically issuing alerts.
In contrast to traditional systems that rely on predetermined definitions or parameters for detecting specific objects, movement patterns, or behaviors, Conntour asserts that its system employs natural and vision language models. This integration provides exceptional flexibility and ease of use. For instance, a user could pose a query like, “Find instances of someone in sneakers passing a bag in the lobby,” and Conntour’s system would rapidly scan all recorded or live video streams to deliver pertinent results.
Because the platform incorporates advanced AI models, users can simply ask questions about the video footage and receive text-based answers, accompanied by the relevant video segments, and even generate comprehensive incident reports.
A key differentiator for the company, however, is its scalability. Goldner elaborated that their platform stands apart from other AI video search services due to its design for efficient scaling across systems comprising thousands of camera feeds. He further noted that Conntour’s system can oversee up to 50 camera feeds utilizing a single consumer-grade GPU, such as Nvidia’s RTX 4090.
Conntour achieves this efficiency by integrating multiple models and logic systems. Its algorithm intelligently determines which models and systems are most appropriate for each query, thereby minimizing the computational power required to deliver optimal results to users.
Conntour states that its system offers flexible deployment options: entirely on-premises, completely cloud-based, or a hybrid combination. It can seamlessly integrate with most existing security infrastructures or function as a standalone surveillance platform.
Nonetheless, the video surveillance industry has long grappled with a fundamental challenge: the effectiveness of surveillance is directly proportional to the quality of the captured footage. For instance, extracting clear details from video recorded by a low-resolution camera with a smudged lens in a dimly lit parking lot remains exceedingly difficult.
Goldner explains that Conntour addresses this inherent limitation by assigning a confidence score to its search results. If the incoming camera feed quality is suboptimal, the system will indicate this by returning results with lower confidence levels.
Looking ahead, Goldner identifies the primary technical hurdle as fully integrating advanced Large Language Model (LLM) capabilities into their system while simultaneously preserving its current high level of efficiency.
“We are simultaneously pursuing two objectives that inherently conflict. On one hand, we aim to provide comprehensive natural language flexibility, similar to LLMs, allowing users to ask anything. On the other hand, there’s the crucial aspect of efficiency; we want to minimize resource consumption, because processing thousands of feeds is an immense undertaking. This fundamental contradiction represents the most significant technical obstacle and problem in our field, and it is what we are diligently working to resolve.”