General Sense was founded on the conviction that artificial intelligence has been built with a sensory blind spot, and that closing it is the most undervalued opportunity of the AI era. The progress of the last decade has run on two data modalities — language and vision — and both encode the world as humans happen to experience it. Cameras and microphones tell a machine what is visible or audible. Neither tells it what is in the air, in the water, on a person's breath, inside a shipping container, or growing in the soil.
The right way to think about what comes next is by analogy to what has already happened with vision. Computer vision and the models built on top of it have given machines a working physics engine of the real world: depth, motion, occlusion, object permanence, and scene structure, all decoded from photons. Autonomous vehicles, robotics, and the new generation of physical-world AI systems run on that engine today. None of those systems have a comparable engine for chemistry. The molecular world — which determines what is toxic, what is degrading, what is being trafficked, what is incipiently sick, and what is being chemically synthesized in plain sight — remains absent from every machine pipeline in operation.
The company is building the first commercial system that decodes biological olfaction at scale and translates it into structured, machine-readable data. Our long-term ambition is to develop a sensory reasoning model — a foundation model for chemistry that learns the ontology of the molecular world the way large language models have learned the ontology of human language.
Together, the sensor network and the model form what we call the chemistry engine of the real world: a continuously updated, model-readable representation of the chemical state of any environment. Vision gave machines a physics engine; chemistry is the engine that has not yet been built.
Every leap in AI capability has followed the arrival of a new data modality at scale: text gave us large language models, images gave us multimodal systems, and trajectories and actions are now giving us robotics. The frontier labs that have driven this progress are openly searching for what comes next, because the scaling curves on text and vision are bending. The field needs new data, new modalities, and new ways to ground intelligence in physical reality.
Real world chemistry is unclaimed. The physical world is, at its most fundamental level, molecular, and so are most of the threats that define this century — synthetic opioids that chemists design faster than legislators can ban them, biological agents engineered in single rooms, explosives at concentrations no electronic instrument can resolve, cancers that emit volatile signatures years before they are imaged. The visible world is well-sensed; the molecular world remains almost entirely unsensed. Machine intelligence cannot reason about what it cannot perceive.
If chemistry becomes a first-class data modality for machine intelligence, several things become possible at once. Continuous, ambient, sub-parts-per-trillion detection becomes feasible outside the laboratory, in the places where it actually matters. Diagnostics shift from episodic to constant. Biosecurity inherits a sense it has never had. Customs, defense, agriculture, food safety, and environmental monitoring all gain a measurement regime that today is gated by cost, latency, and the rarity of skilled operators.
The deeper implication is for machine reasoning. A model that has learned chemical signatures at scale begins to develop sensory semantics — the capacity to identify objects, processes, and intent from molecular composition alone. A nurse who has just removed nail polish and a chemist assembling a peroxide-based explosive leave overlapping chemical traces; distinguishing them is a reasoning problem, not a detection problem, and it is the kind of reasoning any system operating in the real world will eventually be required to perform.
But why not robots? We do not believe biological olfaction itself will be transplanted into autonomous machines anytime soon — the sensor is biological, and biology is hard to copy. But the knowledge extracted from it — what concentrations co-occur, what patterns predict what outcomes, what chemistry indicates what intent — will reshape how robots, autonomous systems, and software agents understand the environments they operate in and the tasks they are asked to perform.
The sensor that performs this task at the required sensitivity, speed, and adaptability already exists, and it is not a machine. It is the canine olfactory system, refined over forty thousand years of co-evolution with humans and already operating in our streets, ports, hospitals, and homes. The dog is not the theoretically optimal chemical sensor. It is the best sensor that is already integrated into our way of life, and that distinction is decisive: the deployment problem is already solved.
What is missing is the read-out. Conventional canine workflows depend on individual handlers, individual animals, and behaviors that take months to shape and do not transfer between dogs. General Sense is replacing that bottleneck with a Nose-Computer Interface (NCI) — a neural implant that records directly from the olfactory pathway, decodes the chemical world the animal is perceiving, and streams it as structured data into a continuously improving model. Each animal in the network contributes to a shared chemical foundation, and each model improvement propagates back. What an isolated dog learns in months, the network learns once.
Neural technology has crossed the threshold from experimental to deployable. The same advances driving Brain-Computer Interfaces (BCIs) for paralysis and blindness are now available to us, and we are not gated by the clinical timelines that constrain human applications.
Animals already working in security, defense, and diagnostic settings represent a live, continuously generating chemical dataset that no other organization is positioned to collect. The longer the network operates, the wider the moat.
The near-term work is mixture classification and cross-animal generalization. The medium-term work is a self-supervised chemical foundation model, trained on a dataset that exists nowhere else.
The first wave of revenue comes from augmenting and replacing deployed detection programs in defense and security, where the economic case is already understood and the budgets already exist. Instead of spending money to collect data, we make money doing it by solving real problems.
Sensory reasoning is a new discipline, and we are assembling the team that will define it — neuroscientists, AI researchers, hardware engineers, and operators willing to build a category from scratch.