Conntour

**Product and problem.** Conntour builds a natural-language search-and-alerting layer that sits on top of an organization's existing camera network and makes it queryable the way a search engine makes documents queryable. Conventional video analytics -- the "video content analysis" (VCA) built into most video-management systems (VMS) -- can only detect a fixed menu of pre-programmed categories (a person, a car, line-crossing, loitering) that an integrator configured in advance. Anything outside that menu is invisible until someone manually scrubs hours of footage. Conntour's stated premise is that this rigidity is the core failure of physical-security surveillance: operators are drowning in cameras they cannot meaningfully watch. The platform lets a user type free-form queries such as "find a man with a tattoo on his left arm," "a van with a print of fruits on it," or "notify me if someone climbs over the wall," and it searches across live feeds and archives without any preset parameters. Public materials describe three modes -- historical search over recorded video, real-time alerting on described scenarios, and structured data extraction (for example, "how many pickup trucks entered yesterday?"). The company claims a single operator can effectively supervise thousands of cameras at once, and that investigators can review days of footage "in minutes rather than days."

**Core technology and how it works.** Conntour applies modern vision-language modeling -- the same class of multimodal AI that connects images and text -- to the physical-security domain, so that an operator's plain-English description is matched against the visual content of frames rather than against a rigid taxonomy of classes. The distinguishing engineering choices, per the company, are that the system understands "complex, context-rich queries within the context of real-world environments," operates in both live-alert and historical-search modes, and is designed to run *fully on-premises* and to be compatible with "any camera at any scale." On-premises deployment is a deliberate design decision for the target buyers: government, border, and critical-infrastructure customers frequently cannot export surveillance video to a third-party cloud for legal, sovereignty, or operational-security reasons, so an air-gappable, camera-agnostic architecture removes a major adoption barrier and differentiates Conntour from cloud-first competitors. The company publishes operational-improvement figures -- up to a 90% reduction in manual review time, up to an 80% reduction in missed security events, and up to a 70% decrease in false-alarm rates -- which are vendor-reported and should be treated as directional marketing claims pending independent validation, but which correctly identify the three metrics (analyst labor, missed events, false alarms) on which physical-security operations are actually judged.

**Market, customers, and go-to-market.** Conntour targets critical-infrastructure protection, border security, large public-venue monitoring, and intelligence/homeland-security operations -- buyers who already own dense camera estates and are constrained by the number of human eyes available to watch them. The go-to-market is enterprise and government direct sales, reinforced by two accelerant channels unusual for a company of its age: selection into the first cohort of Palantir's Startup Fellowship (a distribution and credibility channel into defense and government accounts) and Y Combinator's network. The most important commercial proof point is that Singapore's homeland-security apparatus became a paying customer within months of launch, per the company and reporting by Calcalist/CTech -- a reference in a sophisticated, security-conscious government market that is disproportionately valuable at seed stage. The addressable market is large and established: physical-security video surveillance and video-analytics software is a multi-billion-dollar global category, and the natural-language/"agentic" search wave is currently repricing it, but that also means Conntour is entering a crowded field rather than an empty one.

**Traction, funding, and third-party validation.** Conntour launched publicly out of stealth on 26 March 2026 with a $7 million seed round led by General Catalyst, with participation from Y Combinator, SV Angel, Liquid 2 Ventures, and others. The company went through Y Combinator's Winter 2025 batch and employs roughly 14-15 people across offices in Tel Aviv-Yafo, Israel, and Miami, USA. Beyond capital, the validation stack is notable for the stage: a tier-one generalist VC lead (General Catalyst), the Palantir fellowship, an early paying government customer (Singapore), and an advisory relationship with Bob Flores, former Chief Technology Officer of the CIA, who is quoted endorsing the plain-English querying capability. Each of these is an independent, reputationally-costly signal that partly offsets the thinness of a seed-stage financial and customer record.

**Founders and team.** Conntour was founded in 2024 by Matan Goldner (CEO) and Tomer Kulla (CTO), both computer-vision specialists. The founding insight is unusually well-grounded in the problem: both founders served as IDF field observers in reserve combat units after 7 October 2023, an experience that directly exposed them to the operational failure of watching many camera feeds under pressure. The broader team is reported to include computer-vision engineers from Israel's Unit 8200 signals-intelligence corps and former Israeli intelligence (Mossad) personnel, giving Conntour both the technical depth for multimodal vision AI and the domain fluency for government and defense sales. This founder-market fit -- builders who were themselves the frustrated end-users -- is a recurring pattern in the strongest Israeli dual-use companies.

**Competitive dynamics.** Conntour competes in a dense and fast-moving arena. Incumbent Israeli video-analytics vendors such as BriefCam (now part of Milestone/Canon), Corsight AI, and Oosto (formerly AnyVision) approach the problem through face/attribute recognition and rule-based analytics; a wave of AI-native entrants -- including US players like Ambient.ai, Coram AI, and Verkada, and Israeli-founded Lumana -- are converging on the same "just search your cameras" value proposition. Conntour's differentiation rests on: (1) *fully open-ended natural-language querying* rather than a fixed detection menu; (2) an *on-premises, camera-agnostic* architecture that fits sovereign and classified environments where cloud-first rivals cannot go; (3) a *government-first reference base* (Singapore, Palantir fellowship, ex-CIA advisor) that is hard for consumer/commercial-first competitors to replicate; and (4) elite intelligence-community pedigree in the team. The countervailing reality is that the underlying vision-language capability is advancing industry-wide, several better-capitalized competitors are chasing the identical thesis, and it is not yet demonstrated that Conntour's accuracy, latency, and integration advantages are durable moats rather than a temporary lead.

**Defense, security, and resilience dual-use.** Conntour sits squarely in the dual-use core rather than at its adjacency. The same product that lets a shopping-mall or stadium operator find a lost child or a suspicious package is what lets a homeland-security agency monitor a border, a critical-infrastructure operator watch a substation or port perimeter, or an intelligence unit triage days of surveillance in minutes. The company markets explicitly to border security, critical infrastructure, public safety, and intelligence/homeland-security missions, and its first disclosed paying customer is a national homeland-security organization. The on-premises design is itself a resilience feature: it keeps sensitive surveillance data inside sovereign, potentially air-gapped networks. The principal calibration point is that Conntour is a force-multiplier for *sensing and situational awareness* -- it makes existing camera infrastructure vastly more usable -- rather than a weapons or effects system; its strategic value is in intelligence, surveillance, and reconnaissance (ISR) and physical-security operations.

**Stage, trajectory, and key diligence risks.** Conntour is an early, seed-stage company (founded 2024, out of stealth March 2026, ~15 employees, $7M raised) with an unusually strong signal-to-capital ratio for its age. The plausible trajectory is a Series A within 12-24 months predicated on converting the Singapore reference and Palantir channel into a repeatable pipeline of government and critical-infrastructure accounts, then broadening into commercial enterprise security. The key diligence risks are: (1) **crowded, well-funded competition** converging on the identical natural-language-video thesis, several rivals with more capital; (2) **defensibility** -- whether Conntour's edge is a durable data/accuracy moat or a transient lead over commoditizing vision-language models; (3) **customer concentration and long procurement cycles** -- one flagship government customer today, and government sales are slow and lumpy; (4) **accuracy, bias, and false-positive liability** in high-consequence security settings, where a missed or wrongly-flagged event carries real cost; (5) **privacy, civil-liberties, and regulatory exposure** attaching to any mass-surveillance search tool, which can constrain Western commercial and municipal adoption; and (6) **scaling the elite team** and on-premises deployment/integration engineering across heterogeneous camera estates. None is disqualifying at seed, but each materially shapes the risk-adjusted view.