Install VAOS™, run an agent against your webcam, and watch the perception → action loop in real time.
Python 3.10+ on Linux/macOS. Linux ARM is the production target; macOS works for development.
pip install vaos vaos --version
Spin up the runtime against camera index 0 — VAOS auto-detects backend (Rockchip, Jetson, CPU, etc.).
vaos run --camera 0 --hardware auto
# Open the local dashboard at http://localhost:8800
Agents receive Scene objects and emit actions. Five lines of code, full pipeline.
from vaos import Runtime, Agent runtime = Runtime(camera=0) async def on_scene(scene): if scene.confidence > 0.85: print(f"VAOS sees: {scene.summary}") runtime.attach(Agent(on_scene=on_scene)) runtime.start()
Open the dashboard or curl the API to see the structured output. Full schema →
{
"scene_id": "cam-entrance-20260507T091830Z",
"objects": [
{ "type": "person", "confidence": 0.97 }
],
"event": { "type": "presence", "severity": "low" },
"sensor_reliability": 0.93,
"metrics": { "latency_ms": 38 }
}
Browse the plugin marketplace, then install one to extend perception, reasoning, or action.
vaos plugin install ppe-detector vaos plugin install mqtt-publisher vaos run --camera 0 --plugins ppe-detector,mqtt-publisher
Pull the pre-built image for your SoC, mount your camera, run as a service.
# Rockchip RK3588 docker pull vaos/edge:0.9.2-rk3588 docker run -d --device /dev/video0 vaos/edge:0.9.2-rk3588 # NVIDIA Jetson Orin docker pull vaos/edge:0.9.2-jetson