Trainable AI vision. Inside your facility, inside your systems.
CAM-AI turns camera streams into a confidence score for every category you train. It runs on a server in your network and hands its results to the systems you already use.
From camera image to confidence score.
Every frame is filtered, evaluated and scored in real time. What happens with the score is decided by your teams and systems.
- CamerasExisting or new IP cameras, aligned to what matters.
- CAM-AI serveron premises01 · FILTERMotion and reflection filtering02 · MODELAI model trained on your categories03 · SCOREConfidence per category, 0 to 1
- Plugins & APIResults in the format your systems speak.
- Your systemsYour rules decide what follows.
Trained on your images. Refined with your team.
A model learns the categories that matter in your process: the correct state, the errors and everything in between. Your team can review and tag results in the browser, positive and negative, and every confirmed image can flow into the next training.
Years of know-how in every training run.
Learning rate, augmentation intensity, gamma correction and fine-tuning depth are set before each run and adapted while it runs. Our control algorithm steers them by the validation loss, so models converge fast and stay stable, without a data-science team on your side.
Every run is documented in the training dashboard, with its curves and metrics. Shown here: a real run of about 400 epochs, redrawn; each curve on its own scale.
Results where your process needs them.
CAM-AI is built to be connected. Plugins pass every result into your control, MES and alarm systems, in the format they already understand.
{
"camera": "harness-line-2",
"timestamp": "2026-10-01T14:32:07.412Z",
"model": "connector-check-v7",
"results": {
"connector_mated": 0.0213,
"connector_partially_mated": 0.9871,
"connector_missing": 0.0004,
"hand_in_view": 0.0011,
"empty": 0.0002
}
}Plugins: ours or yours
We build the plugin for your system, or your integrators extend the open core themselves. Both work, and both use the same interfaces.
Rules tailored to your process
Combine categories, cameras and time windows into conditions, and choose what follows: a notification, an entry in your database, an action in your process.
Your rules, your thresholds. CAM-AI measures; your process decides.
Runs in your facility. Under your control.
Training and detection run on a server inside your network. Production images can stay where they are made, and your IT keeps control over every connection.
- Training on site
- New models are trained on the same server, with your images, inside your network.
- Remote access on your terms
- Updates and training through a secure connection your IT provides and approves, such as a VPN or remote-access gateway. Nothing else.
- Modular hardware
- A standard server with NVIDIA GPU, sized to your cameras and models. Scales with more GPUs or more servers.
- European cloud on demand
- For a quick proof of concept or where you prefer it: training and evaluation on our servers in Europe.
- Offline operation
- Where your network requires it, CAM-AI also runs without an internet connection.
If a camera can see it, CAM-AI can learn it.
From standard IP cameras to industrial cameras with precision optics. We advise on camera, lens and position, even in tight machine spaces, and work with the cameras you already have.
- RTSP
- RTMP
- HTTP / MJPEG
- ONVIF
- PoE
- LAN
- WiFi
- The cameras you already haveInstalled IP cameras can be used, as long as they see what matters.
- Industrial cameras & precision opticsHigh resolution, zoom and macro optics for small parts and fine details.
- Low light & infraredNight and infrared images are trained like any other image.
Open at the core. No lock-in.
The CAM-AI core is open source under GPL-3.0. Your IT can audit what runs in its network, and you are never tied to a black box. Machine-specific plugins, integration and managed services are commercial.
- AI
- TensorFlow, Keras, EfficientNetV2
- GPU
- NVIDIA CUDA
- Application
- Python, Django, Channels
- Data
- MariaDB, Redis
- System
- Debian Linux
- Interface
- Browser-based, any device
Your data, your rules.
Installed in your network, CAM-AI works under your IT and data-protection policies. You configure what is seen, stored and shared.
- MaskingMark any image area: exclude it from detection, or focus detection on it.
- PeopleDetection and storage can be switched off for people in the image.
- Roles & rightsUser groups with individually configurable permissions.
- EncryptionEncryption of stored images, configurable to your requirements.
Need a specific security or privacy set-up? Tell us your requirements. Configuring CAM-AI to them is part of every project.
You can detect what you can see.
Want to check whether CAM-AI fits your systems? Talk directly to the people who build it.
Request a demo