vision-agent

Why we included this project

VisionAgent, now deprecated by LandingAI, is still worth a look as a worked example of an agentic layer sitting on top of standard vision models. You give it a prompt and an image, and it picks a suitable detector or tracker, then writes runnable Python code plus a test to check the result. The bundled vision tools may be the more durable part: functions like object detection and counting are callable directly from your own scripts, with no need to go through the agent. Since the project is no longer maintained, treat it as a pattern to study or a source of ready-made wrappers, not a base for new production work. If you are weighing agent-driven approaches to computer vision, the architecture write-up from LandingAI is still worth reading.