Private, local-first image and video annotation with on-device AI.
AnnotateIt is a computer-vision annotation tool for building training datasets on your own device. Projects, media, labels and annotations are stored locally, and every annotation model runs on your hardware — there is no account and no dataset upload.
Open the web app · Try a demo project · Download native apps · Read the documentation · Community and support
AnnotateIt supports images and video in projects for:
- Object detection
- Instance segmentation
- Keypoint detection
- Single-label, multi-label and hierarchical classification
The annotator includes bounding boxes, circles, polygons, open polylines, pixel masks and keypoints, and it can render and edit rotated boxes imported from compatible data. Video objects can be represented as tracks: place keyframes and AnnotateIt interpolates the shapes between them while preserving object identity.
AI-assisted annotation and search run locally. The current model and tool set includes:
- MobileSAM, SAM 2.1 Tiny / Small / Large and SAM 3 Tracker for interactive segmentation
- CLIP and SigLIP 2 for text-image matching, zero-shot classification and semantic search
- Grounding DINO Tiny for open-vocabulary detection from text
- RTMPose-m / l / x for COCO-17 pose assistance
- Detection Assistant, Visual Prompt and batch pre-labelling with a review queue
SAM 3 Tracker is exposed as an interactive segmentation engine; it is not automatic video tracking. Custom ONNX import is experimental and supports compatible YOLOv8 detection, segmentation and pose models, RT-DETR, DETR and image classifiers. The import wizard inspects the model, maps labels and tests it before saving.
The local database also provides workflow features normally associated with a server:
- Immutable dataset versions with compare, download and restore
- On-device quality scans for media, labels, geometry, duplicates, class balance and video coverage
- Deterministic train / validation / test splits with seeds, optional stratification, overrides and locks
- Append-only activity history and dataset copies
- A built-in image editor for crop, resize, rotate, flip and colour adjustments
- Portable project archives and full local-profile backups
Annotated imports are supported for COCO, YOLO, Pascal VOC and Datumaro; plain image archives can also be imported as media. Export adds task-specific image and video formats:
- COCO
- YOLO
- Pascal VOC
- Datumaro
- MOT
- MOTS
- KITTI
- Supervisely Video
- Plain ZIP
Some formats cannot represent every task or shape. MOT, MOTS, KITTI and Supervisely Video are export-oriented workflows, while an AnnotateIt-produced annotated archive can restore its app-specific data through the included sidecar. The export flow reports exclusions and lossy conversions before creating an archive; see the format documentation for the exact matrix.
| Platform | Get it | Notes |
|---|---|---|
| Web | app.annotateit.ai | Opens without an install; models download on demand and are cached locally |
| Windows | Microsoft Store | Native full-feature build with bundled engines and the local REST API |
| macOS | Mac App Store | Native full-feature build for Apple silicon Macs |
| iPhone and iPad | App Store | Touch-first build with the complete annotation workflow and a reduced on-device model set |
Every platform is free to use, with no account, trial period, subscription or project limit. Windows and macOS also offer an optional one-time Pro upgrade for multiple workspaces and priority email support; annotation features remain available without it.
The web app downloads its code and any optional model files you choose, but it does not upload your datasets. Native builds bundle the default engines and can run the annotation workflow fully offline.
The only optional product feature that talks to a third party is Ask AI. It stays off until you add your own OpenAI API key. When enabled, your chat messages and, when relevant to the question, project and label names are sent to OpenAI; images, videos and annotations are not.
annotateit-community is the public space for everything that needs a public trail:
- Report a bug
- Request a feature
- Report an import or export problem
- Read the public roadmap
- Read the web app changelog
Please keep private datasets, user images, credentials and personal information out of public issues. Report security vulnerabilities privately to umno.annotateit@gmail.com, never in an issue.
AnnotateIt is proprietary software. The application source code is not published in this organization's repositories; the public repositories contain the community issue tracker, product documentation, roadmap and organization policies.
- Product support: umno.annotateit@gmail.com
- Security reports: umno.annotateit@gmail.com — mark the subject as a security report