A spin-off of the Politecnico di Torino. We build vision models light enough to run right next to your cameras, without sacrificing accuracy.
Focoos AI was founded in 2022 as a spin-off of the Politecnico di Torino by Antonio Tavera, Fabio Cermelli, Barbara Caputo, Vittorio Di Tomaso, Carlo Masone and Giuseppe Averta, turning years of academic research on efficient deep learning into a company.
An early €335K pre-seed round in May 2023 gave the company the means to move from published papers to a real product, and to build its first core team.
Focoos AI was incubated by I3P, the Politecnico's incubator, and selected for the ESA BIC Turin programme, which helped validate the technology on demanding, real-world problems.
Anyma, Focoos AI's Neural Architecture Search engine, designs efficient models automatically, alongside a family of detection and segmentation models, each in four sizes from Nano to Large.
Fast, accurate family trained on COCO and Objects365. Ideal for smart cameras, surveillance and robotics.
Pixel-precise family trained on ADE20K. Ideal for smart cities, self-driving and aerospace.
Tuned for speed on edge hardware without giving up accuracy.
A €2.65M seed round in January 2025, led by Galaxia, with Exor Ventures, Vertis, VC Partners, PiemonteNext and Vento, to scale the company's technology and bring efficient vision AI to industry.
Focoos AI launched a no-code platform to train, test, deploy and monitor vision models, plus a developer library for engineers who prefer to work straight from their own code.
Today Focoos AI is fully focused on Sentinel: a smart virtual operator for your cameras, built on vision language models, that understands what it sees in real-time and runs on the hardware you already own.
Efficient vision AI that runs where your cameras are: accurate, affordable and private by design.
A vision language model reads behaviour and context, not fixed rules, so you get far fewer false alarms.
Runs at the edge, from an NVIDIA Jetson to a server, with no cloud round-trip.
Light models keep running costs low as you scale across cameras and sites.
On-device processing means footage never has to leave your site. Privacy by design.
Events are ranked by urgency and the noise is filtered out.
Safety, intrusion, theft, vandalism and more come preconfigured, so you start on day one.
The best way to understand what we do is to see Sentinel in action.
"*" indicates required fields
We’ll get back to you shortly
