NeuronBox

**Product and the concrete problem it solves.** NeuronBox is developing an AI operating layer for industrial wastewater treatment, beginning with the part of the plant where small process errors can become expensive compliance events: dissolved-air flotation and related chemical-dosing workflows. The company's first named agent, Maya, is presented as an industrial wastewater expert that monitors a treatment line, interprets process conditions, helps operators understand what is happening, and can adjust operating decisions in real time. The problem is concrete. Factories, food processors, dairy plants, and other industrial facilities must keep effluent within permit limits while dealing with variable inflows, changing chemical demand, aging equipment, laboratory delays, and a shortage of experienced operators. A plant may have sensors and a SCADA screen but still depend on a small number of people who know how a particular line behaves under unusual loads. NeuronBox's proposition is to capture that operational expertise in software that works continuously, reduces avoidable chemical spend, and makes compliance less dependent on a single shift supervisor. The official product site emphasizes 24/7 monitoring, reporting, and insight; the public company and market profiles add autonomous process optimization and compliance support. The record should not overstate the current product: public material describes a platform and pilot or design-partner evidence, not a mature fleet of fully autonomous plants.

**Core technology and how it works.** The public technical description is unusually specific for a young company. NeuronBox says Maya integrates with existing SCADA and PLC infrastructure, uses computer vision, reinforcement learning with trajectory prediction, and an agentic orchestration layer, and can actuate through Modbus TCP with sub-second response. PLANETech describes an edge AI box that connects to existing sensors and PLCs, predicts laboratory-grade compliance results in real time, adjusts chemical dosing, and performs root-cause analysis. In practical terms, that architecture appears to combine three layers: observation of process variables and possibly visual plant state; a learned or model-based estimate of how current conditions will affect separation and effluent quality; and a bounded action layer that recommends or executes control changes. The important architectural choice is deployment alongside installed equipment rather than a rip-and-replace treatment system. That can shorten integration if the connector, permissions, and safety envelope are robust. It also creates hard engineering obligations. The system must distinguish sensor drift from a real process change, understand delayed laboratory measurements, preserve operator overrides, respect dosing and discharge limits, and fail safely when communications or data quality deteriorate. The public sources do not disclose model architecture, training-set size, validation protocol, cybersecurity controls, or which actions remain human-approved, so those are central diligence questions rather than assumed capabilities.

**Market, customers, and go-to-market.** NeuronBox targets industrial facilities where wastewater is a production constraint and an environmental obligation. The Israel Innovation Authority identifies factories, food-processing plants, dairy plants, wastewater-treatment operators, water utilities, and environmental or compliance teams as target users. This is a sensible beachhead because industrial plants often have repeatable processes, expensive chemicals, measurable effluent outcomes, and a direct economic reason to reduce operator workload and compliance risk. The initial buyer could be a plant manager, environmental-compliance leader, engineering group, or water-treatment service provider rather than a municipal utility. The public go-to-market model has two visible channels. NeuronBox presents a software or edge-enabled AI workforce that can deploy against existing infrastructure, while Greenfield Eco describes Green Pipe, developed by NeuronBox, as an AI virtual worker for industrial wastewater and identifies itself as the exclusive distributor in Israel. That relationship could let NeuronBox reach plants through an established engineering and wastewater-services workflow rather than selling a new control product directly to every factory. The company also advertises a low-friction pilot and PLANETech describes a SaaS model deployed in days. The commercial test is whether a fast installation can become a repeatable, secure deployment across different PLC vendors, treatment chemistries, and regulatory regimes, instead of turning into bespoke consulting for each plant.

**Traction, funding, and third-party validation.** The public record supports an early validation story with several independent ecosystem touchpoints. The Israel Innovation Authority lists NEURON BOX LTD as an Israeli company established in 2025, with five employees, R&D-stage status, and support under the 2026 Ideation or Startup Fund route; it names Ofer Hayut as CEO and founder and classifies the technology as artificial intelligence, agentic AI, and edge AI for water applications. NeuronBox's company profile on LinkedIn states that Maya has delivered a 2.3-times chemical-efficiency improvement and a 15-times compliance improvement, while PLANETech describes those results as demonstrated at a dairy-factory design partner. Those figures are important signals but remain company or ecosystem reported: the reviewed sources do not publish the baseline, study duration, definition of compliance improvement, sample size, independent auditor, or customer name. Greenfield Eco provides a separate commercial signal by describing a Green Pipe product, offering a pilot, and naming itself as the exclusive Israeli distributor. A public Israeli company-information listing corroborates the 2025 incorporation and active status. There is no publicly verified priced equity round, disclosed revenue, installed-unit count, or customer retention metric in the sources reviewed. The correct conclusion is that NeuronBox has a real legal and operating footprint, a named product, a channel relationship, and promising early performance claims, but not yet the evidence of scaled commercial traction that would justify a mature-stage classification.

**Founders and team background.** Ofer Hayut is publicly identified as NeuronBox's CEO and founder, and his LinkedIn profile associates him with the company and with the Israeli water and environmental-technology ecosystem. The public company profile currently exposes only a small team, while the Innovation Authority lists five employees. That team size is consistent with a young product company still building its first industrial agent, but it is a material constraint for a safety-sensitive deployment platform. NeuronBox needs more than model development. It needs process-control engineering, wastewater chemistry, industrial networking, edge-device reliability, cybersecurity, regulatory interpretation, customer implementation, and technical support. The public record does not establish the full leadership bench, the identity of a CTO, the number of wastewater process specialists, or whether Greenfield Eco supplies part of the domain and deployment capacity. The founder's product framing is strong because it starts with a bounded industrial workflow and measurable outcomes rather than an abstract general-purpose agent. The diligence gap is execution depth: the company must prove that it can turn knowledge of a DAF process into a repeatable product that remains safe when plant conditions, instrumentation, and local permit requirements differ. Until the technical and commercial team is more visible, the team score should remain below the technology and strategic-fit scores.

**Competitive dynamics.** NeuronBox competes against a stack of incumbents and substitutes. Veolia and SUEZ bring end-to-end industrial water services, treatment operations, chemical expertise, and long-standing customer relationships. Xylem and Evoqua combine water-treatment equipment, analytics, and service contracts, while Siemens and Schneider Electric can provide PLC, SCADA, edge-computing, and industrial-automation layers into which a plant may already be standardized. Specialized process-control and instrumentation vendors such as Yokogawa and Hach compete through monitoring, measurement, and automation rather than through a fully autonomous wastewater agent. The internal operations team and conventional engineering integrator are also direct substitutes: many factories prefer a human operator plus fixed dosing rules, periodic lab testing, and a service contract. NeuronBox's potential edge is the combination of domain-specific AI workers, existing-infrastructure integration, rapid deployment, and an action loop that aims to connect prediction to process control. A named DAF specialist such as Maya could make the system more useful than a generic industrial chatbot because it is organized around chemistry, flotation dynamics, dosing, and compliance. The edge is not yet a durable moat. Incumbents own installed control points and data, a large vendor can add AI features to a service contract, and every new plant may require process-specific tuning. Defensibility will depend on proprietary operational data, safe-control policies, deployment speed, validated outcome improvements, and trust earned through repeated plant references.

**Defense, security, and resilience relevance.** NeuronBox is not publicly presented as a defense contractor, and no reviewed source identifies a military customer, classified program, defense contract, or fielded security deployment. Its dual-use relevance is nevertheless credible through industrial and critical-infrastructure resilience. Wastewater treatment supports food factories, hospitals, energy and chemical facilities, military bases, and municipal services; failure can force a production shutdown, create an environmental incident, or remove a local water-reuse capability. An edge-capable system that can continue monitoring and assisting operators during staffing shortages, network degradation, or a disruption to remote expertise could improve continuity. The same architecture may be relevant to emergency water-treatment operations, distributed industrial sites, and facilities that must keep running while supply chains or specialist access are impaired. It also has a cyber-physical security dimension: connecting AI to PLCs and dosing systems requires strong authentication, network segmentation, signed updates, audit logs, human authorization, and a clear safe state. Those controls are not optional extras when the product can influence physical treatment processes. The appropriate assessment is therefore strategic resilience dual use, not direct defense capability. NeuronBox's relevance would rise materially if it demonstrates secure edge deployment, offline or degraded-connectivity operation, auditable operator approval, and successful continuity testing in critical facilities; it should not be credited for those properties before they are publicly evidenced.

**Growth stage, trajectory, and key diligence risks.** NeuronBox is best classified as early. It has moved beyond an idea: it has a named agent, an official product site, a government innovation record, a distribution relationship, a market-profiled design partner, and a public claim of measurable process improvement. It remains early because the company was established in 2025, has a very small disclosed team, is listed at R&D stage, and has not publicly shown a priced financing round, revenue, deployment count, independent validation, or regulatory and security certification. The plausible trajectory is to start with DAF and chemical-dosing assistance, expand Maya into adjacent industrial water workflows, and then reuse the underlying agent architecture for other process industries. The principal diligence points are: (1) reproduce the 2.3-times chemical-efficiency and 15-times compliance claims against a documented baseline; (2) verify the dairy design-partner scope and distinguish pilot from paid production deployment; (3) map every autonomous action, limit, override, and fail-safe; (4) test connectors across PLC, SCADA, sensor, laboratory, and historian environments; (5) establish cybersecurity, tenant isolation, data residency, patching, and incident-response controls; (6) determine whether the agent generalizes beyond one DAF configuration; and (7) confirm funding runway, intellectual-property ownership, channel economics, and the support burden per site. Risks include noisy or delayed measurements, process-specific tuning, liability for incorrect dosing, long industrial sales cycles, incumbent bundling, small-team execution, and the possibility that AI features become standard components of larger water-treatment platforms. The company merits monitoring because it sits at the intersection of AI infrastructure, industrial autonomy, water security, and operational resilience, while the evidence still calls for disciplined technical and customer diligence.