CNX Software dropped the spec sheet on April 20 and the embedded crowd hasn’t shut up about it. Two SKUs share one IP67 enclosure: the CQ38W-1126B runs Rockchip’s RV1126B with a 3 TOPS NPU, the CQ38W-3576 swaps in an octa-core RK3576J/M with 6 TOPS. The big one ships for $149.
Why this is a VLM-camera moment
6 TOPS in a sealed industrial cam is enough headroom for YOLO detection plus a 2B-parameter LLM — Qwen, Gemma2-2B, Phi2, MiniCPM all listed as supported. That’s not a chip benchmark, that’s a camera that thinks about what it sees in a parking lot or on a forklift. Ring and Hikvision still ship everything to the consumer cloud. Firefly is handing ODMs the price anchor for the next generation of cameras that talk to agents instead of apps.
How an agent plugs in
TensorFlow, ONNX, PyTorch, and Caffe models deploy straight onto the device. Standard RTSP and ONVIF push the stream and inference results out, so a Home Assistant or LangGraph agent can subscribe — VLM polls a snapshot, decides if a pallet is stacked wrong, fires a tool call. No cloud round-trip, no SDK lock-in. The monitoring camera becomes an eye your agent can read.
You Might Also Like
- Onecli Puts a Security Gateway Between Your ai Agents and Your api Keys
- Alibaba Qwen Smart Glasses g1 s1 275 ai Glasses With Swappable Batteries and a Qwen api Backend
- Raspberry pi ai hat 2 130 Gets you a Local llm box With 40 Tops and its own 8gb ram
- Ring app Store Amazon Opens its 100m Camera Fleet to Third Party Apps
- Qwen Image 2 0 Just Dropped and i Honestly Wasnt Expecting This
Top AI Product 
Leave a comment