Partners & investors
Work with AnnotateIt
Two kinds of conversation are welcome here: with teams that run a computer vision pipeline and want a say in what the tool does next, and with investors who want to talk to the person building it. Both reach the same person.
Who we want to work with
The teams that fit are small, technical and already annotating — and the work usually looks like one of these four situations.
A model pre-labels; people verify
A YOLO, SAM or detector checkpoint produces the first pass and your engineers correct it. AnnotateIt runs that loop on the machine that holds the data — SAM 2.1 and SAM 3, Grounding DINO text prompts, RT-DETRv2, DEIM, RF-DETR and ECDet detection checkpoints, RTMPose for human pose, and your own YOLOv8 ONNX export imported as a custom model. Every proposal lands in the same editor as a hand-drawn shape, reviewable and easy to throw away.
Models that run on-deviceVideo that has to become a dataset
Sports footage, vessel CCTV, drone flights, advertising creative. Extract frames at an interval or by hand, label intervals with ranges, place keyframes on tracks that interpolate between them, and export frames and annotations in a format the training script already reads.
Video to datasetLarge image sets, a small team
Camera traps, inspection imagery, field surveys: tens of thousands of images and one or two people reviewing detections. Batch auto-annotation, deterministic train/val/test splits, dataset versions and quality checks are all in the free tier — no project limit, no per-seat cost.
Batch auto-annotationA pipeline that calls the tool instead of clicking it
An internal platform that needs to push media in and pull annotations out. The Windows and macOS builds can expose a REST API bound to the loopback address — projects, datasets, media, annotations and video tracks over HTTP on the same machine, off until you turn it on, with no hosted service in between.
The local REST API
Annotating on CVAT today? Export the project as COCO, YOLO or Datumaro and import that — the standard formats are auto-detected. CVAT XML itself is not an import format, and we would rather say so here than let you find out on the import screen.Dataset formats
What a design partnership involves
Nothing formal. No contract to sign and no data to hand over — your media and annotations stay on your machines, as they do for every user. What changes is the distance between you and the code.
What you bring
- A real pipeline with real data and a real deadline — the kind that exposes what a demo dataset never does.
- Forty minutes to walk through it: which model runs first, who corrects what, where the export goes.
- Blunt feedback when something is slower, clumsier or missing compared with the tool you use today.
What you get
- The developer who writes the code on the other end of the thread — not a ticket queue.
- Your exports and your models checked against the actual product before you move anything: the COCO or YOLO round trip, the custom ONNX import, the video frames.
- The gaps you hit moved ahead of the generic items on the public roadmap, and early builds when they are fixed.
- A written case study only if you want one, with your approval on every sentence — and nothing published about you without asking.
Where it is not a fit — yet
The second list on the About page applies here too, and three items on it come up in exactly the conversations this page invites.
- Several annotators sharing one queue, with assignment and review stages. There is no server and no account system, so there is nothing to share a queue on.
- 3D cuboids, LiDAR sequences and sensor fusion. AnnotateIt has an experimental PLY/PCD point-cloud semantic editor, without these workflows.
- A labelling workforce. There are no annotators for hire here; the product is for the people who already have the domain knowledge.
If one of these is your blocker and you would rather help define what a local-first answer to it looks like than wait for one, that is a conversation worth having — say so in the first email.
What AnnotateIt is — and is notFor investors
AnnotateIt is bootstrapped and built by one person. The product is shipped, not planned: a web app, native Windows and macOS builds, and iPhone and iPad builds, currently free on every supported platform. Release notes are published on this site and the roadmap is public.
The thesis is narrow. Annotation tooling has been built around the cloud service for a decade, while the models that do the heavy lifting — segmentation, detection, pose — now run on the laptop that holds the data. A local-first tool with those models inside it serves the individual engineer and the small technical team, the people the cloud platforms were never sized for.
We are open to conversations with angel and strategic investors who know computer vision tooling or the teams that use it. This page states no terms and is not an offer of any kind; it is an invitation to talk, and the first reply comes from the founder.
Start the conversation
An email is enough. If you would rather see the product before writing, the web app opens a sample project without an account.