No-Code AI Vision

A first-pass no-code computer vision workflow builder, similar in spirit to n8n but focused on camera pipelines.

The included sample workflow is:

Camera Loader -> Object Detection -> Class Filter -> Live Preview -> Alert Output

The app runs locally and uses the browser camera APIs. Object detection is powered by TensorFlow.js COCO-SSD loaded from a CDN, so the first model load needs internet access.

Run

.\.venv\Scripts\python.exe main.py

Open:

http://127.0.0.1:8000

Camera access works on localhost/127.0.0.1 in modern browsers. Click Run, grant camera permission, then adjust the nodes in the inspector.

Current nodes

• Camera Loader: chooses camera, resolution, and facing mode.
• Object Detection: selects COCO-SSD model variant, confidence threshold, and inference interval.
• Class Filter: passes only configured classes such as person, car, dog.
• Live Preview: draws bounding boxes and labels.
• Alert Output: writes detection events with a cooldown.

Next useful steps

• Add persisted workflow import.
• Add RTSP/IP camera loading through a backend worker.
• Add backend inference nodes for YOLO/ONNX/TensorRT.
• Add webhook, database, and snapshot output nodes.