Agentic computer vision
The Swiss knife for computer vision.
Label your images, train a model without code, test it honestly and download it for your device. The assistant plans the work with you, explains the results and proposes what to do next. Nothing to install, your data stored in Switzerland and the EU.
Accounts are opened by Pixtell. Join tells us about your project, and Pixtell answers within two working days.
In the box
- An assistant that sets up the plan with you
- Labeling in the browser
- Smart data selection
- No-code training
- No infrastructure to set up
- Synthetic image generation
- Every build recorded
- Edge and cloud deployment
How a project goes
- 1
Brief
Write a few sentences about what to find.
- 2
Plan
The assistant proposes a plan. You approve it.
- 3
Data
Add your images from an upload, a bucket or CVAT.
- 4
Suggested labelsComing soon
Label a few. Few‑shot and zero‑shot suggest the rest.
- 5
Build
Vision Foundry trains the model in the cloud.
- 6
Model
Vision Foundry tests it on images kept aside and never trained on.
- 7
Evaluation
Read the verdict against your bar.
Repeat until the model meets your barComing soon
High-value data selection: Vision Foundry picks the images worth labeling next. Coming soon
- 8
Deployment
Approve the model and download its package for a device or a server.
Everything a vision project needs
From the first image to a model on your device, without code and without other tools to buy.
An assistant that sets up the plan with you
Write what you want to find. The assistant asks what it needs, proposes a plan and adjusts it with you.
Labeling in the browser
Draw boxes, outlines and points on your images, and see at a glance which labels look wrong.
Smart data selection
Choose the images that matter: saved selections across your datasets, a check of the labels and a map of every object.
No-code training
Once you approve the plan, training and testing run by themselves. No code, no settings to guess.
No infrastructure to set up
Graphics cards in the cloud and storage for every image, included. Nothing to rent or install.
Synthetic image generation
Vision Foundry generates training images to fill the gaps in yours.
Every build recorded
Each build keeps its plan, its training set and its scores, so two models compare on the same test images.
Edge and cloud deployment
Download the approved model as a package that runs on a device at the edge or on a server in the cloud.
Your images and labels are stored in Zurich and the EU. Pixtell is the one company you deal with.
Questions
Who is Vision Foundry for?
Teams that have images and something to find in them. No machine learning engineer and no infrastructure needed.
What does agentic mean here?
The assistant works out the right workflow with you, explains the results and proposes what to do next. You approve it before it runs, and you can see every step of it in the workflow builder.
What kinds of models can I build?
Models that tag each image (classification), find objects with boxes, outline them (segmentation) or place points on them (poses).
Do I need to install anything?
No. Everything runs in your browser and in Pixtell's cloud.
Where is my data stored?
In Switzerland and the EU.
What do I get at the end?
A package with the model in the ONNX format, which runs on a device or a server. For some devices, such as an NVIDIA Jetson, Pixtell also provides a ready-made container that converts the model to run fast on that device.
How do I get an account?
Through Join. Pixtell answers within two working days.
Who owns my data?
You do. Your images, labels and datasets stay yours, and Pixtell uses them only to provide the service to you, never to train models for others without your consent. Terms, section 4
Where does the training happen?
On graphics cards that Pixtell provides in the cloud. There is nothing for you to rent.
Have images and something to find in them?
Tell Pixtell about your project. Pixtell answers within two working days.