Geo-sleuth Turns Any Coding Agent Into an Open-Source Photo Detective
Geo-sleuth is an open-source agent skill that locates photos to within meters using OpenStreetMap, elevation skylines, satellite tiles and street view, with no landmarks required.
- Geo-sleuth is an open-source agent skill that locates photos with no GPS or landmarks
- Works with Claude Code, Codex, Cursor, Gemini CLI, OpenCode and GitHub Copilot via a standard SKILL.md
- Combines OpenStreetMap geometry, elevation skylines, CLIP satellite tile ranking and DINOv2 street view matching
- Documented cases hit 2 m, 87 m and 3 m from the true camera position in 32 to 72 minutes
- Scripts rank candidates and the model only judges the top few, with every claim tied to a file
- MIT licensed, needs Python 3.10+, uv, curl and Chrome or Playwright Chromium
Geo-sleuth turns coding agents into photo geolocators
Geo-sleuth is an open-source skill for Claude Code, Codex, Cursor, Gemini CLI, OpenCode, and GitHub Copilot. Given a photo without GPS metadata, readable signs, license plates, or recognizable landmarks, it estimates the camera’s location and heading, then generates a satellite view and supporting evidence. The repository has attracted more than 3,200 GitHub stars.
The project formulates geolocation as a geometry and data-search problem. Pier spacing can estimate distance, shadows can constrain camera direction, and a ridge line can be compared with terrain models. Scripts search and rank candidate locations using public datasets; the language model reviews a short list and cites artifacts generated during the run.
Scripts do the searching
Geo-sleuth follows the Agent Skills format, with a SKILL.md instruction file and roughly 20 single-purpose Python scripts. Its tools cover six main sources of evidence:
| Signal | Method | Purpose |
|---|---|---|
| Metadata and text | EXIF extraction, OCR through Apple Vision or RapidOCR, and reverse image search through Baidu and Yandex | Find embedded coordinates, camera details, visible words, or matching images |
| Map geometry | OpenStreetMap queries through Overpass | Locate roads, railways, power lines, bridges, and other linear features |
| Terrain | Skyline rendering and ridge matching with AWS Terrain Tiles | Compare visible mountain profiles with elevation data |
| Satellite imagery | CLIP zero-shot scoring | Rank tiles containing features such as running tracks, factories, or silos |
| Street-level imagery | DINOv2 similarity and SIFT feature matching |
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
