A two-year-old startup just closed a $7 million seed round in 72 hours to build what it calls a Google-like search engine for security video feeds. Conntour, co-founded by CEO Matan Goldner, pulled in backing from General Catalyst, Y Combinator, SV Angel, and Liquid 2 Ventures, TechCrunch AI reports in an exclusive interview.
Goldner scheduled roughly 90 investor meetings in eight days. By Wednesday afternoon of the first week, the round was done.
🔍 What Conntour actually does
The platform lets security teams query camera footage using plain language. Think of it as natural language search, but pointed at surveillance feeds instead of the web.
A user can type something like “Find instances of someone in sneakers passing a bag in the lobby,” and the system searches recorded or live video to return matching clips. It can also monitor feeds in real time, detect threats based on preset rules, and surface alerts automatically.
What makes this different from legacy video analytics? Traditional systems rely on predefined parameters: specific objects, motion patterns, behaviors you’ve already told the system to look for. Conntour uses vision-language models, which means users can ask open-ended questions they haven’t pre-programmed.
The system can also answer questions about footage in text form and generate incident reports on the fly.
⚡ The scalability angle
This is where Conntour’s pitch gets interesting. According to Goldner, the platform can monitor up to 50 camera feeds off a single consumer GPU like Nvidia’s RTX 4090. For context, most AI video processing systems chew through compute resources fast: scaling to thousands of cameras typically requires serious infrastructure.
Conntour manages this by routing queries through different models and logic systems, picking whichever combination requires the least computing power while still delivering accurate results. The system deploys on-premises, in the cloud, or hybrid, and plugs into most existing security setups.
🎯 The ethics question
This launch comes at a loaded moment for surveillance tech. Flock’s camera network is under fire for ICE surveillance partnerships. Ring is drawing criticism for features that let law enforcement request neighborhood footage from homeowners. The privacy debate is loud.
Goldner told TechCrunch AI that Conntour is “quite picky” about its clients. The startup already counts Singapore’s Central Narcotics Bureau and several large government and publicly listed companies among its customers.
“We’re really in control of who is using it, what is the use case, and we can select what we think is moral and, of course, legal,” he said.
Whether selective client screening holds up as the company scales with fresh capital is a fair question. Startups under growth pressure from investors don’t always maintain that discipline.
🧠 The hard problem ahead
Goldner is candid about Conntour’s biggest technical challenge: full LLM-level flexibility versus efficiency. These two goals directly contradict each other.
“On one hand, we want to provide full natural language flexibility, LLM-style, to let you ask anything. And on the other hand there’s efficiency, so we want to make it use very few resources, because processing thousands of feeds is just insane,” he said.
The company also addresses a fundamental limitation of video surveillance: bad footage. Poorly lit areas, low-res cameras, dirty lenses. Conntour returns a confidence score with results, flagging when source quality is too low for reliable answers.
📌 Why this matters
Vision-language models have matured fast. What was a research demo two years ago is now powering production systems that process thousands of live camera feeds. Conntour’s seed round, closed in three days from investors like General Catalyst and YC, signals strong conviction that AI-powered video search is ready for enterprise scale.
The surveillance AI market is growing whether privacy advocates like it or not. The more relevant question is who builds the guardrails. Conntour is betting that selective customer screening and transparency around confidence scores are part of the answer.
More details are available in the original report on TechCrunch AI.