Emberwatch pairs a real time YOLO26 detector with NVIDIA's DAM-3B vision model and an NVIDIA reasoning model to turn a camera feed into a structured fire and smoke safety advisory.
A fast detector finds it, a vision model describes it, a reasoning model decides.
Use your camera, or upload an image or short clip.
Click Start and allow camera access,
or upload a file below.
YOLO finds the regions, NVIDIA DAM-3B describes them, and the reasoning model returns a safety advisory.
No advisory yet. It appears after a sustained detection on the live feed, or right after you analyze a file.
First run can take 30 to 60 seconds while the GPU model wakes up.
No alerts yet.
You have reached the daily free limit of . Enter an access token to continue.
Optional. A token removes the daily free limit.
Optional: leave an email so we can reach you. You can skip this.
A hybrid pipeline for facility safety. YOLO26 finds fire and smoke in each frame. On a sustained detection, NVIDIA DAM-3B describes the exact region and an NVIDIA reasoning model returns a structured safety advisory.
Live or upload. Start your camera, or upload an image or short clip.
Detection. YOLO marks fire and smoke boxes in real time.
Advisory. You get severity, affected zone, recommended actions, and an escalation level.
Heads up: the first advisory after idle can take 30 to 60 seconds while the GPU model wakes up. After that it is fast.
Flare stacks, steam, and vapor can look like incidents; the model weighs that and may flag a false alarm.