Private AI
We run large language and vision-language models on dedicated hardware in our own lab, exposed through standard OpenAI-compatible APIs. Everything a model reads — documents, images, conversations — stays inside the local network.
Incanet s.r.l. — Research and Development Labs
We study emerging technologies before they become products.
An independent lab exploring applied artificial intelligence, immersive technologies and privacy by design. We build working prototypes, run them on our own infrastructure and learn from what breaks.
01 — Artificial Intelligence
Intelligence that runs where the data lives.
Our AI research focuses on models and agents that work on local hardware, inside the network that owns the data. No cloud dependency, no telemetry, full control over what the model sees.
Computer vision
Real-time analysis of video streams: motion detection, scene description, event classification over time. We study how a model can understand what happens in a space from a few frames — and when it is better to stay silent.
Agents and automation
Agents that observe event streams — sensors, cameras, messages — reason about them with a language model and decide whether to act, notify or wait. We research de-duplication, rule-based pre-classification and the boundary between a useful alert and noise.
Conversational interfaces
Voice in real time, 3D avatars with lip sync, assistants embedded in devices and in virtual spaces. We study how people talk to a machine that has a face, and how much latency a conversation can tolerate.
02 — Other research areas
Where the lab also looks.
Virtual reality and 3D
Immersive environments, WebXR, digital twins of physical spaces and 3D capture techniques such as Gaussian splatting.
Secure communication
Encrypted channels, self-hosted messaging and mail infrastructure, VPN topologies that keep a network reachable without exposing it.
Privacy by design
Architectures that minimise collected data from the start: no third-party trackers, no external fonts, no analytics. This website is built the same way.
Applied research for companies
When a company brings us a problem, we treat it as a research question: prototype first, measure, then decide whether it deserves to become a product.
03 — Approach
Research, prototype, learn.
-
Research
We read, test and compare before writing a line of code. Open source first, open standards whenever possible.
-
Prototype
Working software over slides. Every idea becomes something that runs on real hardware, in a real network, with real data.
-
Learn
We run our prototypes every day in our own spaces. What survives daily use is what we talk about.
04 — Lab notes
What is running in the lab right now.
Internal projects, in daily use. No client data, no third-party services involved.
Local inference server
A dedicated GPU machine serving large vision-language models on the local network through a standard API, shared by every other experiment in the lab.
Camera understanding pipeline
4K streams from several cameras, frame-difference motion detection, parallel workers sending wide and cropped views of the same instant to the vision model, episodes tracked from start to end.
Event classification agents
A real-time pipeline that collects events from the local message bus, classifies them with the local model after de-duplication and rule-based pre-filtering, and notifies only when it matters.
Embodied assistant
A 3D avatar in the browser with real-time voice, lip sync and AR/VR mode, connected to a self-hosted agent backend.
05 — Contact
Let's talk about a research question.
We are a small lab and we choose our collaborations carefully. Write to us with a problem, not a brief.
info@incanet.it