Running AI inference entirely on-device with no cloud dependency is the core pitch of the 8devices 8Sight T100, a compact long-wave infrared camera built around STMicroelectronics' STM32N6 Arm Cortex-M55 microcontroller. The STM32N6's Neural-ART accelerator delivers up to 600 GOPS of NPU compute, enough to handle object detection, anomaly classification, and temperature-triggered events at the edge with AI inference times under 100 ms. Paired with a Lynred ATI320 LWIR sensor capturing 320x240 thermal video at 60 fps and sensitivity below 50 mK, the T100 fits into a housing measuring just 28.5 x 30.5 x 29.5 mm (1.1 x 1.2 x 1.2 in) and weighing 38 g (1.3 oz), purpose-built for drone and robotics integration.
The STM32N6 has been gaining traction in the open-source embedded AI space. STMicroelectronics publishes object detection starter code on GitHub, and Edge Impulse maintains open-source firmware for the platform. Other STM32N6-based camera boards like the CamThink NeoEyes NE301 ship with open hardware files and open-source firmware. The 8Sight T100 takes a different path: its firmware is proprietary, and 8devices does not currently publish an SDK, API reference, sample code, or model deployment workflow. Custom firmware development is available as a paid service, and integration with the company's RobonodeVision system and Robosoft Linux platform is expected in early access during Q4 2026.
Connectivity covers both USB 2.0 and 100 Mbps Ethernet via JST connectors. The USB interface supports UVC-compliant streams in MJPEG, H.264, and YUV422 alongside raw formats, while the Ethernet port handles MJPEG and H.264 over RTP, RTSP, and RTCP. Onboard resources include 32 MB of RAM and 64 MB of NOR flash, with four M2 mounting holes for flexible physical integration.
8devices has opened an early access program for the 8Sight T100, with initial engineering samples expected to ship by the end of 2026 and a first production run planned for Q1 2027. Pricing has not been announced.
