Since April 2026
AI vision is still too frame-based, compute-heavy, and inefficient for a world that moves continuously.
Security cameras, self-driving cars, AR glasses, robots, drones, healthcare systems, manufacturing lines, smart homes, and edge devices need to understand motion, attention, and prediction in real time — not just recognise objects in single frames.
At the same time, AI’s energy cost is becoming a major bottleneck. More capability often means larger models, more GPUs, bigger data centres, and higher power consumption.
Okko builds visual AI inspired by how biological vision handles continuous change.
Instead of relying on ever-larger models and heavier compute, Okko creates a more efficient visual architecture that focuses on motion, attention, and the most important parts of a scene.
We start in software to prove the concept, then use the validated design to develop analogue chips for low-power vision in cameras, robots, AR glasses, cars, drones, and edge devices.
Visual AI is moving from image recognition to always-on video understanding, but current models are becoming too large, expensive, and energy-hungry for real-world devices.
Cameras are now everywhere — in cars, factories, drones, robots, homes, hospitals, and AR glasses — and they need real-time perception that is fast and efficient without relying on massive data centres and compute.
Biological vision is one of the most efficient computing systems in nature. A medium banana contains about 105 kcal, enough energy to support roughly six hours of whole-brain activity at about 20 watts, while animals use that kind of energy to navigate, track motion, avoid danger, and understand the world continuously.
AI is moving in the opposite direction. More capability often means more GPUs, larger data centres, and rising electricity demand; the IEA projects global data-centre electricity consumption to roughly double to around 945 TWh by 2030. Yet despite this energy cost, current AI still struggles with video: many systems process sampled frames, take seconds or longer to interpret short clips, and often miss the structure that matters most — motion, order, speed, attention, and what happens next.
Okko starts from that gap. Nature shows that continuous visual understanding can be efficient, adaptive, and always on. Okko begins by testing these principles in software, then uses the validated architecture to guide low-power analogue chips for cameras, robots, AR glasses, cars, drones, and edge devices.
Okko is currently in the early product-development stage, focused on proving the core architecture in software before moving toward hardware.
We have built the first version of the visual input system: a biologically inspired software retina that converts images into structured receptor activations. This gives us the foundation for testing how visual information should be sampled, compressed, and passed into later layers of the model.
The current work is focused on expanding this into a broader visual architecture inspired by biological vision, while keeping the implementation modular enough to test, measure, and improve each stage. We are validating the concept in software first, with the long-term goal of translating the strongest parts of the architecture into low-power analogue chip designs.
Vision
Okko’s long-term vision is to build efficient perception systems for the physical world.
We start with vision because cameras are already everywhere, but current AI is too heavy for many real-time, always-on applications. Our goal is to move from software prototypes to dedicated analogue chips that can run visual intelligence directly on edge devices — plugged into computers, camera systems, AR glasses, autonomous vehicles, drones, robots, and industrial equipment.
The first generation of Okko chips would act as a low-power visual front end: processing motion, attention, and scene structure close to the sensor, before information ever needs to reach a larger model or data centre. Over time, this could make visual AI faster, cheaper, and more practical in the real world.
Vision is only the first step. The same biological approach can later expand into sound, touch, proprioception, and other sensory systems. Okko’s goal is to build the perception layer for intelligent machines — and eventually enter robotics with systems that do not just compute, but sense the world efficiently.
Competition
Okko sits near event-based vision and neuromorphic companies such as Prophesee, iniVation, SynSense, and BrainChip. Prophesee and iniVation focus on event-based cameras and neuromorphic vision sensors that capture changes in light with low latency, high dynamic range, and less redundant data than normal cameras. SynSense and BrainChip focus more on neuromorphic chips and low-power edge AI processing.
These companies show that the market is moving toward faster, more efficient visual systems, but much of the category remains limited to sensing, event capture, or early-stage retinal-style processing.
Okko’s difference is that we are not building an isolated visual sensor or a system that stops at the first few layers of the retina. We are building toward a fuller perception architecture: one that does not only capture visual change, but processes it into motion, attention, scene structure, and eventually conceptual understanding.