Overview AI released the OV Spark line of edge AI inspection cameras with Sparky, an onboard AI agent that handles tool selection, diagnostics, and operator interface generation through natural language interaction. The cameras ship with 10 inspection tools—seven deep learning models and three traditional vision checks—that can be combined in a single inspection recipe. All processing runs locally on the camera.
The technical hook is adaptive alignment. The camera anchors inspection regions to part features rather than fixed coordinates, allowing it to track parts that shift or rotate in robot cells. If no part is detected, the capture fails rather than passing an empty frame—a safeguard against false positives in automated lines.
Sparky addresses a workflow problem Overview AI identified after seven years of factory deployments: tool selection and data interpretation still required specialist judgment, creating setup bottlenecks. The agent recommends which tools an inspection needs and explains the reasoning. It also makes the capture library searchable in plain English, so operators can query failure patterns by timeframe or defect type without writing SQL or parsing logs.
Mansfield Engineered Components deployed the first production unit on an end-of-line station that previously relied on manual visual checks. A quality engineer with minimal vision system experience built the inspection—targeting missing washers as the primary defect—and has since moved the camera between two lines. MEC is scoping a second unit for rivet curl and stenciled date code verification.
The 1.2MP OV Spark and 5MP OV Spark Pro are built on Qualcomm Dragonwing silicon, run entirely at the edge, and measure 70 x 116 x 53 mm with IP66 sealing. Both models are available now through Overview AI and its distribution partners.



