Psistar

**Product and the concrete problem it solves.** Psistar addresses the gap between the large volume of telemetry collected by critical facilities and the operational judgment needed when those facilities behave unexpectedly. A power plant, data center, aerospace system, oil-and-gas asset, or defense platform can expose pressure, flow, temperature, vibration, electrical, and control data while the most valuable operating knowledge remains distributed across P&IDs, control logic diagrams, alarm lists, manuals, standard operating procedures, and the experience of a few senior operators. Conventional dashboards show more signals but do not necessarily tell a control-room team which physical state is developing, why it is developing, or which safe action should follow. Psistar's stated product thesis is an agentic team member for high-stakes operations: it combines a continuous sensor stream with the customer's playbook and returns predictions, explanations, and operational recommendations at the point of work. The company explicitly targets environments where connectivity may be constrained and where an incorrect recommendation can carry safety, availability, or mission consequences. This is an operational-reliability and decision-support product, not a generic chatbot layered over industrial data.

**Core technology and how it works.** The company describes two model families. The first is a neuro-symbolic reasoning engine that converts engineering documentation into a queryable operational knowledge graph with traceability back to source material. Psistar's newer public descriptions call this structured representation a Plant Model: physical assets and relationships, sensor or tag semantics, logical dependencies, operating rules, interlocks, and procedures are represented so that an AI system can reason over the facility rather than retrieve isolated passages. The second is a physics-informed forecasting model, referred to in public ecosystem material as Horizon, that predicts the future state of dynamic systems from sensor telemetry and is intended to support zero-shot deployment to assets it has not previously seen. The official site says the models treat pressure, flow, temperature, and related physical variables as physical tokens, then classify predicted states against manuals, protocols, and schematics. Psistar positions the system as edge-native, offline, air-gapped, and auditable, with large language models limited to an interface or orchestration layer rather than placed in the final decision path. Public sources do not disclose benchmark datasets, error rates, model weights, or the exact boundary between deterministic logic, statistical forecasting, and human approval, so those claims require technical validation.

**Market, customers, and go-to-market.** Psistar's initial market is the operational technology and industrial-AI layer beneath sectors where downtime, unsafe conditions, or loss of expert knowledge are expensive. The official site names power plants, aerospace, defense, data centers, and oil and gas; the Israel Innovation Authority record names defense as the target sector, while IVC describes high-stakes operational environments and a broader industrial customer set. The product can be introduced without rewiring a facility or adding new physical sensors because the company says it consumes existing sensor streams and technical documentation. That creates a plausible software-first deployment motion: begin with a reliability or anomaly-detection proof of value on one asset, ingest the relevant engineering corpus, demonstrate forecast and explanation quality, then expand across a site or fleet. Buyers could include heads of operations, reliability engineering, plant managers, data-center infrastructure teams, defense-prime integrators, and government operators. The commercial opportunity is attractive because the same architecture can address predictive maintenance, energy efficiency, operator training, incident response, and resilience. The hard part is enterprise integration and trust: customers will demand connectors for industrial protocols, clear liability boundaries, change-control procedures, cybersecurity reviews, and evidence that recommendations improve availability without increasing operational risk.

**Traction, funding, and third-party validation.** Psistar is young but has several independent signals beyond a company landing page. The Israel Innovation Authority lists PSISTAR LTD as an Israeli company established in 2025, with ten employees, an R&D stage, and two in-house model families spanning engineering-document reasoning and physics-informed forecasting. Startup Nation Finder reports a March 2025 founding date, a 1-10 employee range, and a $1.9 million seed round in December 2025 led by ICI Fund. ICI's own portfolio page describes Psistar as agentic operations and maintenance for mission-critical environments, combining telemetry, operational know-how, anomaly prediction, and explainable recommendations on air-gapped systems. IVC lists the company as formerly Maverick AI and reports that it was seeking capital, which supports an early financing and commercialization posture but should not be confused with a closed additional round. Globes included Psistar among the Israeli companies presented at VivaTech 2026, describing it as building physics-informed foundation models. The team also publicly promoted a selection among 40 companies for a Tel Aviv regional final of The Pitch by Deel. These are meaningful ecosystem and fundraising signals, but there is no public revenue figure, named production customer, quantified uptime improvement, or independently audited model evaluation in the sources reviewed.

**Founders and team background.** Public records identify Yochai Pagi as Psistar's co-founder and CEO and Matan Pagi as CTO and co-founder. Yochai's public founder profile describes prior service as a major in the Israeli Air Force, a Magshimim graduate background, and a computer-science master's degree; his public writing frames the company around the practical failure of language-model AI in physical, noisy, high-consequence environments. That background gives the company a credible connection to operational systems and defense requirements, even though the sources do not establish a specific military unit, classified program, or defense contract. Matan Pagi is identified consistently in the Israel Innovation Authority record and company ecosystem material as the technical co-founder, but detailed independent biography and prior employer information are not publicly established. Psistar's public careers page indicates a small engineering organization working across physics or time-series modeling, graph and vector retrieval, production AI pipelines, FastAPI and WebSocket services, and a React and TypeScript operator interface. It also mentions industrial protocols such as Modbus and OPC UA, air-gapped deployment, evaluation, regression suites, and benchmark discipline. The disclosed ten-person scale and still-forming technical team are both an advantage for focus and a diligence risk for delivery, support, and domain coverage.

**Competitive dynamics and edge.** Psistar competes with several established approaches rather than entering an empty category. Siemens and Senseye combine industrial asset management, predictive maintenance, and digital-twin capabilities; AVEVA and OSIsoft-derived PI System deployments own deep industrial data and historian relationships; Cognite Data Fusion provides industrial context and data integration; C3 AI sells predictive-maintenance and operations applications; and Augury focuses on machine-health sensing and diagnostics. Defense and infrastructure customers may also compare the product with Palantir Foundry and AIP, which provide data integration, workflow, and AI layers for mission operations, even though they do not share Psistar's exact physics-native architecture. Psistar's possible edge is the combination of traceable engineering knowledge, a system-level model of the facility, physics-informed forecasting, and air-gapped execution in one layer that can sit on existing infrastructure. That combination could reduce the cost of adapting AI to a new plant or asset and make outputs more acceptable to safety-critical operators than an opaque general-purpose model. It is not yet a demonstrated moat. Industrial incumbents own installed data, control-system relationships, and service organizations; specialist vendors may have better machine-level benchmarks; and a customer could assemble much of the stack from a historian, knowledge graph, time-series model, and workflow platform. Psistar must prove lower deployment time, better zero-shot transfer, useful causal explanations, and safe recommendations under sensor drift and missing data.

**Defense, security, and resilience dual-use relevance.** Psistar's dual-use case is structurally credible because the core technology is designed for both commercial industrial operations and defense or national-resilience environments, rather than being a civilian product with a weak security metaphor attached afterward. A power plant or data center needs early warning of abnormal states, operational continuity during degraded connectivity, and a reliable way to preserve scarce expertise; a defense platform, aerospace system, fuel facility, or military logistics site needs the same capabilities under stricter availability, cyber, and air-gap constraints. The official site explicitly lists defense and aerospace and emphasizes offline operation; the Innovation Authority classifies the company under defense-related technology and describes an auditable reliability and optimization layer for existing control systems. Potential deployments include isolated command-support facilities, generator and microgrid fleets, aircraft or propulsion test environments, data centers supporting sovereign workloads, and critical industrial sites where expert operators may be unavailable during a crisis. The public record does not show fielded military capability, a government contract, safety certification, or a classified deployment. The strategic relevance should therefore be calibrated as enabling infrastructure for resilient operations and defense-industrial readiness, with strong adjacency to operational security and continuity but without claiming proven battlefield use.

**Growth stage, trajectory, and key diligence risks.** Psistar is best classified as early stage. It has an active Israeli legal entity, a disclosed seed investment, a ten-person ecosystem estimate, a public product site, a hiring footprint, and external recognition from the Innovation Authority, ICI Fund, Finder, and Israeli technology media. Those signals place it beyond an undisclosed idea, but they do not establish repeatable commercial scale. The most important upside path is to turn a difficult technical integration into a reusable deployment system: ingest engineering documents, construct a traceable plant model, adapt the forecasting layer to new assets, validate forecasts against historical and live conditions, and convert recommendations into approved operator workflows. If that process becomes repeatable, Psistar could become a strategic software layer for facilities where cloud dependence is unacceptable and operational knowledge is a bottleneck. Diligence should focus on: (1) independent forecast and anomaly-detection benchmarks under drift, missing sensors, and regime changes; (2) evidence of customer pilots converting to paid deployments; (3) the safety case, human-approval design, and liability allocation for recommended actions; (4) cyber hardening, data residency, update procedures, and true air-gap support; (5) the accuracy and maintenance cost of the Plant Model; (6) financing runway and the ability of a small team to support industrial deployments; and (7) whether Siemens, AVEVA, Palantir, or control-system vendors can replicate the feature set through installed distribution. The thesis is strategically strong but still proof-driven.