Label anything, natively on your Mac.

Type a word, click once, or show one example and LabelFlow outlines the rest, in this image or across all of them. It runs on your Mac, with Claude when you want it.

Get early access See how it works

Five ways to label. Use whichever is fastest.

A street scene where every car is outlined after typing car

Type it

Type “car” and SAM 3.1 outlines every car in the image, each with how sure it is.

Text prompt

Click it

One click outlines the whole object. Use Apple’s on-device segmenter for speed, or SAM 3.1 for detail.

Interactive

LabelFlow's Find similar: one boxed car, every other car on the street outlined with a confidence

Show one

Box a single example and Find similar marks every match in this image. Set the confidence, then add them all in one go.

Find similar, in this image

LabelFlow's Memory: four people saved as examples on one image, then every person found and outlined on two other images

Teach it once

Save an outline to Memory and LabelFlow learns that object. Then Find in All Images looks through the whole project and proposes it everywhere, for you to accept or reject.

Memory, across images

Drawing a box by hand in the LabelFlow editor

Draw it

Boxes, polygons, circles, lines, points and a mask brush, with the size shown live while you drag.

Manual tools

Claude works inside LabelFlow.

Connect Claude and ask in plain words. Every change lands in the app’s History, where you can see it and undo it.

You

Find a street-scene dataset on Hugging Face, import a hundred images and name the classes.

Claude, through the LabelFlow connector

Done. They’re in a new project called Streets, labelled and ready to review.

Search and import datasets from Hugging Face, Kaggle and Zenodo
Outline objects, label whole folders and learn from your examples
Check labels for problems and export when you’re happy

The LabelFlow editor with Claude's import and rename in the History panel

The details

Your images stay on your Mac

Labelling and the AI run on-device. Nothing is uploaded unless you choose to share it.

Standard formats

Export COCO JSON, YOLO (boxes or segmentation) and Pascal VOC. Labels are saved as labelme JSON next to your images.

Review built in

AI suggestions arrive as proposals you accept or reject, and every shape keeps its source and confidence.

Native on Apple silicon

A Mac app, not a browser tab. Windows, Linux and web versions are in the works.

LabelFlow is in early access.

Join the list and we’ll email you when there’s a place for you.