A local GeoGuessr-for-real-life: it reads EXIF GPS and reasons over visual clues (signage, plates, architecture, flora, sun position) using a local uncensored vision-language model + a reasoning model — no cloud, no API keys, nothing uploaded.

pip install "cognis-locateanything[img]"
fleet up vision reasoning      # via https://github.com/cognis-digital/uncensored-fleet
locate photo.jpg               # → ranked candidates + rationale
locate photo.jpg --format json
locate photo.jpg --exif-only --format geojson   # offline EXIF fix, straight onto a map

Output formats

--format	use
table (default)	human-readable ranked candidates
json	machine-readable, for pipelines / evidence logs
geojson	RFC 7946 FeatureCollection — open in QGIS, Leaflet, Mapbox, geojson.io

--exif-only is a deterministic, offline, model-free run that uses only the embedded EXIF GPS fix — ideal for CI, batch triage, or air-gapped review.

🔎 Example output

Real, reproducible output from the tool — runs offline:

$ locateanything-emit --version
locateanything 0.1.0

$ locateanything-emit --help
usage: locate [-h] [--version] [--format {table,json,geojson}] [--exif-only]
              image

Infer where a photo was taken (local VL + reasoning model).

positional arguments:
  image                 path to an image

options:
  -h, --help            show this help message and exit
  --version             show program's version number and exit
  --format {table,json,geojson}
  --exif-only           offline, deterministic: use only embedded EXIF GPS,
                        skip the VL/reasoning models

Blocks above are real locateanything output — reproduce them from a clone.

Sample result format (illustrative values — run on your own data for real findings):

{
"findings": [
    {
        "id": "1234567890",
        "title": "Suspicious Activity Detected",
        "description": "An unknown actor has been observed attempting to access a sensitive system.",
        "created_at": "2023-02-15T14:30:00Z",
        "updated_at": "2023-02-15T14:30:00Z",
        "labels": ["suspicious", "malware"],
        "indicators": [
            {
                "type": "ip",
                "value": "192.0.2.1"
            },
            {
                "type": "domain",
                "value": "example.com"
            }
        ]
    }
]
}

Usage — step by step

1. Install the CLI (console-script: locate):

   pipx install "git+https://github.com/cognis-digital/locateanything.git"
   locate --version
   

2. Infer where a photo was taken (runs entirely on a local vision + reasoning model):

   locate ./photo.jpg
   

3. Get machine-readable output for pipelines or evidence logs:

   locate ./photo.jpg --format json > location.json
   

4. Read the result — parse the JSON for the inferred location and rationale:

   jq '.' location.json
   

5. In CI/batch, loop over a folder of images and collect findings:

   for f in images/*.jpg; do locate "$f" --format json; done > all_locations.jsonl
   

Demos

Worked, runnable scenarios live in demos/ — each has a SCENARIO.md and, where relevant, a sample image carrying a real public landmark coordinate in EXIF so you can run it end-to-end offline with --exif-only. (Re)generate the sample images with python scripts/make_demo_images.py.

#	Scenario
01	Basic run — full VL + reasoning
02	EXIF GPS fix, offline (no models needed)
03	Batch a folder into JSONL
04	GeoJSON export → drop straight onto a map
05	Southern + Western hemisphere (sign handling)
06	Disaster-response / situational awareness
07	Maritime / port geolocation (suite interop)
08	Evidence chain + forward to STIX/MISP/Slack
09	No EXIF → visual-clue inference

Architecture

flowchart LR
  IMG[📷 image] --> EXIF[EXIF GPS parse]
  IMG --> VL[Uncensored VL model<br/>visual clues]
  EXIF --> R[Reasoning model<br/>rank candidates]
  VL --> R
  R --> OUT[Ranked locations + rationale<br/>table / JSON / GeoJSON / MCP]

Use it from any AI stack

• MCP server (locate mcp) for Claude Desktop / Cursor / uncensored-fleet
• JSON output pipes into any agent · LangChain/CrewAI tool in one line · plain CLI

⚠️ Responsible use

For OSINT, journalism, and research. Get consent before geolocating images of people or private property, and comply with local law. You are responsible for your use.

Related

🤖 uncensored-fleet · 🧠 engram · 🔍 geolens · 🗂️ the suite

⭐ If this is cool, star it — it helps others find it.

Interoperability

locateanything composes with the 300+ tool Cognis suite — JSON in/out and a shared OpenAI-compatible /v1 backbone. See INTEROP.md for the suite map, composition patterns, and reference stacks.

Integrations

Forward locateanything's findings to STIX/MISP/Sigma/Splunk/Elastic/Slack/webhooks via cognis-connect. See INTEGRATIONS.md.

License

COCL v1.0 — see LICENSE.