Intelligence, with instinct.
Inztinct builds ZIV, one autonomous base that carries five robot modules. Every robot runs on three layers of our own: Core, Sense and Learn.
One base. Five jobs.
Every ZIV shares the same drive, power, compute and Inztinct Core. A module docks on top, the base recognises it, and the right Safety profile loads.
Prototype reference dimensions. Subject to change before release.
A drone for the pit's ESG report.
The Inztinct Mining Drone flies a fixed route over open-pit mines and records dust, slope movement, water and rehabilitation, with the same Core, Sense and Learn underneath. It is for compliance evidence, not exploration.
Every robot is built with all three.
Intelligence plans the task. Inztinct decides what the robot is allowed to do about it.
Core
Cannot be talked into anything.
- Fixed, measurable limits
- Its own sensors and drive cut-off
- Never updated over the air
Sense
Knows a covered table from an open one.
- Learned awareness of the surroundings
- Lowers limits to match the place
- Decides on the machine
Learn
Gets better. Never gets looser.
- Federated, offline learning
- Tested, signed releases
- No path to Core
What happens when the AI is wrong, or taken over?
The intelligence can be fooled. Core cannot be talked to. It sits below the AI with its own sensing and a fixed set of limits. Pick a scenario.
Illustrative simulation of the architecture. Not measured data. Limits shown are example values.
Instinct is what you are born with. Intelligence is what you learn.
A robot that works near people needs both. The intelligence plans, adapts and improves. The instinct is fixed, fast and always on, so a bad decision upstairs never reaches the wheels.
The thinking happens on the machine.
A mine with no signal, a corridor with no Wi-Fi, a stop that cannot wait for a server. Inztinct machines decide on board. The cloud is for reports and updates, never for the decision.
Evidence and events go up. Signed releases come down, to Intelligence and Sense only. Core is never updated over the air.
Cameras, lidar, IMU, bumpers
Wheels, steering, payload
Design intent. Not measured data.
No cloud in the safety path
Anything that keeps people safe runs on the machine and keeps working if the network, or the AI above it, does not.
Sealed where it is captured
For the Mining Drone, survey records are dated, placed and signed on the device, before they go anywhere.
Works where there is no signal
Pits, plant rooms and basements. The machine does its job offline and syncs when it can.
We have been writing about this since 2018.
Our founder wrote about intelligence moving to the edge in December 2018, under the name Internet of Intelligent Things. The same thinking later led to Economy of Edge, a machine-payments project. Inztinct applies it to robots that work near people.
Reference compute for the Intelligence layer.
Models and safety interfaces that do not depend on one accelerator.
Our own carrier board, with compute and the safety core on one module. Not designed yet.
NVIDIA and Jetson are trademarks of NVIDIA Corporation. Inztinct is a member of NVIDIA Inception. Membership is not an endorsement or a partnership.
Concept scenes from the ZIV family.
ZIV autonomous base
Aluminium chassis, four drive and steer wheels, a universal payload interface and a Jetson-class compute unit.
ArchitectureInztinct Core, Sense, Learn
Fixed limits, learned awareness and a fleet that improves, inside every robot we build.
Deep techInztinct chip
A staged research path from a safety microcontroller to a purpose-built safety chip.
What we are learning.
Building with NVIDIA: where ZIV and Halos meet
NVIDIA is putting safety at the centre of physical AI. Here is how we see our work fitting beside theirs, with credit where it is due.
8 Oct 2026A ROM protects the code, not the clock
Hardware roots of trust get broken by glitching power. What a Tesla jailbreak teaches us about the limits of any safety chip.
6 Oct 2026Instinct, intuition, intellect: a vocabulary for physical AI
The robot-learning world already talks about System 1 and System 2. We think a fixed layer deserves its own word.
Working with.
Build the robots people can trust near people.
We are talking to pilot partners in logistics and inspection, and to investors ahead of our seed round.