Edge AI devices
Cameras, terminals and modules with an AI model running locally: recognising images, sound and sensor signals without sending raw data to the cloud.
Edge AI and devices
Edge AI: we build devices with built-in artificial intelligence
Not every AI has to live in the cloud. We build devices that process images, sound and sensor data on site: on the shop floor, in a vehicle or in your own server room. From the first consultation, through an electronics prototype, to a working MVP and an AI cluster under your own roof.
65 min technical consultationQuote in writingNo commitment
What is edge AI
Edge AI is artificial intelligence that runs directly on a device, close to where the data is created, instead of sending everything to the cloud. A camera spots a defect on the line by itself, a sensor detects unusual machine vibration by itself, and a terminal understands a voice command by itself. Only the result goes to the network: an alarm, a count or a record in your system. That gives you three things the cloud can't: a decision in a fraction of a second, operation without internet, and data that never leaves the company.
From idea to device
Pick a use case and see the device architecture and the path from consultation to series production. It's a simplified example: every project starts from your own case.
1. Choose a use case
2. Device architecture
- SensorIndustrial camera
- On-site computeAI module at the line
- AI modelDefect detection in the image
- ActionPart rejected + logged in the report
The line doesn't wait for a server. The good/defective decision is made while the part is still under the camera.
3. On the device or in the cloud?
Response timedecision on siteEdge: 5 out of 5.depends on the connectionCloud: 2 out of 5.
Data privacyimages and audio stay putEdge: 5 out of 5.raw data at the providerCloud: 3 out of 5.
Works offlinekeeps runningEdge: 5 out of 5.system stopsCloud: 1 out of 5.
Data transfer costonly the result is sentEdge: 5 out of 5.constant data streamCloud: 1 out of 5.
An indicative rating for this use case, not a measurement. We train the model on images from your production; it then runs on the module at the line.
4. Path to deployment
We check whether the problem can be solved on a device: what data, what response time, in what conditions and in what quantities.
How we build an AI device
Consultation and feasibility
We define what the device should recognise, the conditions it works in and what it connects to. We check whether the task can run at the edge or whether part of it has to go to a server.
Prototype
We choose the compute module, cameras and sensors, and design the electronics and enclosure for the prototype. We train the AI model and optimise it for the specific hardware so it fits its memory and power budget.
MVP and pilot
The device works in a real environment: on the line, in the warehouse, in a vehicle. We measure accuracy, latency and failure rate, and improve the model on your data.
Series and maintenance
We prepare a version ready to replicate, remote model updates and monitoring of the device fleet. Your own team can also take the project over.
What we build
On-premise AI clusters
Your own infrastructure for running language and vision models in your server room. For companies that can't or don't want to send data to outside providers.
Electronics and prototypes
Electronics design, sensor and camera integration, enclosure and device software. We bring our experience in industrial automation and robotics to AI devices.
End-to-end MVP and consulting
We take a project from idea to a working MVP, or advise your team at selected stages: hardware choice, architecture, model optimisation.
Where edge AI makes sense
Edge AI wins wherever response time, unreliable connectivity or data confidentiality matter.
Manufacturing
Inline quality control, part counting, detecting abnormal machine operation from vibration, temperature and sound.
Logistics and warehousing
Reading labels and markings, counting pallets, monitoring forklift safety zones. It works even where Wi-Fi doesn't reach.
Companies with confidential data
Law firms, healthcare, finance: language models on a cluster in your own server room, with no documents sent outside.
Case study: Synapse, IoT on custom electronics
For four years we led the development of Synapse, an IoT system that collects, processes and analyses data from racing yachts on custom electronics with edge computing. The platform integrated our own strain-gauge sensors measuring loads on carbon rigging in real time. It's the same kind of work we now bring to AI devices for industry: sensors, electronics, on-site processing and data that reaches the system only as a result. The project is described in our portfolio.
Edge AI or the cloud
The cloud is a good fit when you need the largest models and fractions of a second don't matter. Edge AI wins when a decision has to be made instantly, the connection is unreliable, streaming images from many cameras would be expensive, or the data can't leave the plant. In practice we often build a hybrid: the device decides on site, and only results and hard cases for analysis go to the server.
An AI cluster in your own server room
Open-weight language models can now run on your own hardware with quality good enough for many business tasks: document search, summaries, classification and assistants. We size the hardware to your users and workloads, set up the cluster, connect it to your systems and keep it updated. Hardware is always quoted separately and in writing.
How much does an AI device cost
Cost depends on what the device has to recognise, how many units you need, the conditions it works in and the systems it connects to. The cheapest way to test an idea is a prototype on off-the-shelf modules, with custom electronics designed only after a pilot. You get a written quote after the consultation.
Frequently asked questions
What is edge AI?
It means running artificial intelligence models directly on a device, for example in a camera, a controller or a small computer next to a machine, instead of in the cloud. Data is processed where it's created.
Does the device work without internet?
Yes, that's one of the main advantages of edge AI. The model runs locally, and results sync with your systems when the connection comes back.
Do you also build the electronics?
Yes. We design electronics and prototypes and integrate cameras and sensors, and for larger runs we work with contract manufacturers.
Can we start with a consultation?
Yes. Many clients start with a technical consultation or a review of an existing project and decide on the build afterwards.
Who owns the device and the model?
Ownership of the code, the model and the electronics design is agreed in the contract before we start. We can hand the project over to your team for further development.
Data security: nine principles we follow on every project.How we protect data
Let's talk about your device
Tell us what it should recognise and where it will work. In 65 minutes we'll assess whether it's a job for edge AI, the cloud or a hybrid, and what the path to a prototype would look like.
Book a consultation