Skip to content

Lannotex

Fast, offline-first image annotation for classifications, bounding boxes, and polygons, on your own machine.

Lannotex is a free, MIT-licensed desktop app. Each project is one SQLite file in a folder you choose. Your source images are never moved, copied, or altered. There is no account, no telemetry, and no network traffic unless you deliberately host or join a team.

Install Lannotex Coming from LabelMe?

Drawing two boxes and a polygon on a cat photo in Lannotex, then saving and moving to the next image

Why Lannotex

  • Projects that scale

    One project.sqlite3 per dataset instead of thousands of JSON files. The image list is virtualized: it keeps at most 1,024 records in memory, whether the dataset has 300 images or 100,000.

  • You don't lose work

    Every edit goes to a recovery draft, autosave runs 700 ms after you stop, and saves are transactional. Invalid geometry, such as self-intersecting polygons or zero-size boxes, is rejected before it reaches your dataset.

  • Every common format, free

    Export to Labelme JSON, YOLO, COCO, classification CSV, or a lossless native JSONL. You can export from the app or headlessly with lannotex export in CI.

  • Optional teams, no cloud

    Host a project on your LAN, invite reviewers with private tokens, and sync in the background. A stale save becomes a conflict you resolve; it never silently overwrites someone else's work.

At a glance
Annotation types Whole-image classes (multi-label), bounding boxes, polygons
Pipelines Classification, detection, segmentation, or combined
Platforms Windows, macOS, Linux; Python 3.11–3.14 or standalone binaries
Exports Labelme JSON, YOLO boxes, COCO JSON, classification CSV, native JSONL
Imports Images (JPEG, PNG, WebP), Labelme JSON, native JSONL
License MIT

Where to next