Civimatica

Civimatica S.r.l.

Physical AI for the real city.

We build AI models for urban space, run them on our own street-level sensing devices, and give companies the data and ground truth to train and validate theirs. Privacy by design, from the first byte.

01Models meet an unmeasured world

AI is leaving the data centre. The street isn't ready.

Models built for physical space are trained on simulations, scraped proxies and occasional surveys, then deployed onto streets that behave nothing like the data that produced them.

A delivery robot validated in simulation meets a kerb, a cyclist and a double-parked van for the first time in production. An urban model trained on payment records learns who paid, not what the street was doing.

The information exists. It is simply unmeasured, and where it is measured, it is usually collected in ways public space cannot accept.

02What Civimatica does

We build the models, the machines, and the ground truth.

Our own sensing devices measure what happens at street level: occupancy, flow, movement, processed on the device itself.

On top of that measured reality we build and operate our own models. And we open the same substrate, datasets, devices and an instrumented real-street environment, to companies building theirs.

Our models
AI for urban space, trained on streets we measure ourselves.
Your models
Data, devices and test runs for teams building their own.

03How it works

Sense. Model. Deploy.

Sense

Purpose-built devices deployed in the street measure occupancy, flow and movement continuously. Real signal, not a yearly survey.

Model

Models train on measured urban state to produce prediction and behaviour: what a corridor does next, how a machine should read a kerb.

Deploy

Our models run at the edge on our own devices. Yours get validated on instrumented routes with measured ground truth.

04Offering: our models

Civimatica Models

AI models for the physical city, trained on streets we measure ourselves. We build and operate them end to end, delivered as APIs and edge deployments, not slideware.

  • Urban state modelsLive, queryable state of streets, kerbs and corridors, built on measurement, not estimates.
  • PredictionWhat a street or kerb will do next, learned from continuous real-world signal.
  • Custom buildsModels built to your spec and your machines, trained on street data that can't be scraped.
Explore Models

05Offering: your models

Civimatica Sandbox

Simulation only takes a model so far. Sandbox is our test environment in the real city: instrumented routes, measured ground truth and curated datasets for teams building autonomous delivery, robotics and urban AI.

  • Physical test environment

    Instrumented real-street routes where autonomous delivery robots and vehicles can run, fail and be measured.

  • Urban datasets

    Curated, continuously refreshed street-level data for training and benchmarking city AI models.

  • Validation runs

    Repeatable scenarios with ground truth, so a model's behaviour in the city can be proven, not asserted.

Explore Sandbox

06Offering: physical AI

Civimatica Devices

The physical layer the models live on. Our own sensing hardware sits in the street and runs inference at the edge, measuring the city without collecting who is in it.

  • Edge inferenceModels run on the device itself. What leaves is already aggregated street state.
  • Built for public spaceHardware designed to measure space and movement, never identities.
  • Operated by usWe deploy, maintain and iterate on the fleet across our instrumented routes.
Explore Devices

07Privacy by design

Understanding a city should not mean surveilling the people in it.

01 · On the street

The device measures space, not people

Our sensors look at the street itself: whether a kerb slot is occupied, how a corridor is flowing, how objects move through space. Frames are processed in place, in memory.

Leaves this step

  • Occupancy state
  • Flow and speed bands
  • Object class (car, bike, van)

Never leaves / never stored

  • Raw images
  • Faces
  • Plates
  • Anything that names a person

No identities at the edge

Our devices are built to measure the state of the street, space, flow, movement, not who is in it.

Aggregate before it leaves

Data is reduced on the device. What travels to our platform is already stripped and aggregated.

European by construction

Built and operated in Europe under EU data rules, with data minimisation as a design constraint rather than a policy page.

08Who it is for

Robotics & autonomy

Teams that need real streets, real edge cases and measurable ground truth before deployment.

Companies building models

Teams that need urban AI trained on reality, built with our data, our devices, or our test runs.

Cities & operators

Administrations and mobility operators that need evidence, not estimates, about their own streets.

09Why now

Three shifts arriving in the same street.

  1. 01

    AI is moving into physical space

    Models are leaving the data centre for streets, robots and vehicles, and physical AI needs physical data that almost nobody measures.

  2. 02

    Autonomy has run out of simulation

    The remaining gap for delivery robots and vehicles is real-world validation in dense, messy urban space.

  3. 03

    Privacy pressure is a moat, not a tax

    European deployment increasingly requires systems that never collect personal data in the first place. Ours does not.

Let's build on the real city.

Investors, robotics teams, enterprises and cities: tell us what your models need to understand about the street, and we will tell you what we can measure.

hello@civimatica.com

Civimatica S.r.l.