AI Pulse · Under the Radar

A new open-source agent skill is aiming to answer a deceptively hard question from a single image: where was this photo taken? It combines OpenStreetMap geometry, elevation data, satellite imagery and street-view evidence, then exposes the process through an MCP tool that works with Claude Code, Codex, Cursor and Gemini CLI. Rather than returning only a location guess, it shows the reasoning and evidence used to get there. You can see the project and its early documentation in the original repository.

That matters because image-based location work comes up more often than most businesses expect. Think property and construction teams checking site photos, insurers reviewing claims, logistics teams validating delivery evidence, journalists verifying user-submitted images, or security teams investigating an incident. Today, this work often depends on a specialist who knows which visual clues to inspect and where to cross-check them. A tool like this could make that investigation more repeatable, while still leaving a human responsible for the final call.

The project reached 103 GitHub stars in its first day, at a reported pace of 72 per day, which suggests strong early developer interest. That is not proof it is accurate enough for operational decisions. Location guesses can be wrong, source imagery can be incomplete, and evidence needs to be retained when the outcome affects money, safety or compliance. But it is a useful example of agents moving beyond text generation into evidence gathering across multiple data sources.

This is the kind of thing we build into an AI command centre, where tools can collect evidence, show their working and route uncertain cases to the right person.