How do you choose between a labeling platform, a labeling workforce and a scripted tool?
Start from who will do the clicking. If your own domain experts will label, you need a platform with roles and review, and Kili Technology is built around that case: an agreement score is measured before data is exported and each version of a judgment is kept with the person who made it.
If your own data scientists will label in code, Prodigy fits better, because its annotation workflows are Python scripts called recipes and nothing needs a server contract. If nobody on your side has time to label, you are buying a workforce, and the shortlist becomes Kognic for driving data, isahit and Innovatiana for general work.
The second question is the data type. Kognic handles camera, LiDAR and radar together. Supervisely lists image, video, LiDAR and DICOM medical volumes. Picsellia, Lightly, SentiSight.ai and Keylabs concentrate on vision (Keylabs adds 3D), while Kili Technology and Prodigy are the two that clearly cover text. A team labeling contracts and support tickets should not shortlist SentiSight.ai, which the pages checked describe as images only.
A third, less visible split is whether labeling is a project or a pipeline. Picsellia sells annotation as one stage ahead of training, deployment and monitoring, and Supervisely adds neural network training. Lightly sits in front of the labeling step and picks which samples deserve a label at all. If you only need boxes drawn on a folder of images this month, those platforms are more than you need.
What do European data labeling platforms cost?
Five of the ten publish something. SentiSight.ai is the cheapest entry: 1 euro per 1,000 labels and 3.6 euro per hour of training, with a 20 euro sign-up credit and 5 euro free every month, which the vendor equates to 5,000 labels a month.
Supervisely has a free Community edition with 5 GB of storage and Pro from 199 euro per month. Prodigy is a one-off purchase at $390 personal or $490 per company seat in packs of 5, each with 12 months of upgrades.
Picsellia sits at the other end. Its Data Engine module is 1,000 euro per month or 12,000 euro per year with three seats, extra users cost 500 euro per month, labeler seats 10 euro per month, and annotation services 9.60 euro per hour. That is a pipeline price, not an annotation price, and a single annotator is better served by Supervisely or SentiSight.ai.
Kili Technology, Kognic, Lightly, Keylabs, isahit and Innovatiana publish no price list. Kili offers a free trial of 2 to 5 seats and 100 text or image assets, Keylabs a free trial, isahit a free small project and Innovatiana a free quote. Budget for a sales conversation, and ask in it what the metering unit is, because per label, per seat, per hour and per project are all in use.
Does choosing a European labeling vendor keep your data in Europe?
A European contracting entity settles which law governs the contract and which authority supervises it. Kili Technology names MYAI SAS in Paris, Kognic names Kognic AB with the Swedish Authority for Privacy Protection as supervisory authority, Picsellia names CNIL as the complaint route, and Lightly AG is registered in Zurich under Swiss law. That part is verifiable on the vendors' own legal pages.
Where the data physically sits is a separate fact, and many vendors leave it open. Picsellia says its SaaS uses EU and US data centres. Kili, Supervisely, SentiSight.ai, Kognic, Keylabs, isahit and Innovatiana do not state a hosting region on the pages we checked. Only a few tools remove the question: Prodigy runs on your own hardware, Kili and Picsellia can run on your own Kubernetes (Picsellia also air-gapped), and Supervisely and Lightly list on-premise deployment.
For workforce vendors there is a third location, the annotators themselves. Isahit says its workers are largely in developing countries and Innovatiana describes itself as French and Malagasy. A European contract does not mean a European pair of eyes on your images, and a data protection agreement should say who may see the raw files.
How hard is it to move from Scale AI to a European alternative?
The software part is the easy part. Scale AI, Inc. is a San Francisco company whose Data Engine is sold through demos, and its own site describes work from annotation to RLHF and red teaming. European platforms take your existing labels in common formats: Lightly lists COCO, YOLO, Pascal VOC and KITTI, and Prodigy exports JSON. Budget days, not months, for moving the tooling.
The hard part is the people. Scale supplies a contributor workforce, and the vendor claims 25 percent hold advanced degrees. If your quality today depends on that workforce, replace it deliberately: Kognic for autonomy data, isahit or Innovatiana for general annotation and RLHF, Kili Technology with your own experts as reviewers. Run a paid pilot on the same 500 items you sent to Scale and compare agreement before cutting over.
Plan for contract terms too. A Scale master services agreement will name its own subprocessors list, while a European vendor will hand you a data processing agreement under GDPR. Ask both sides for deletion of the old labeled sets once the transfer is verified.
When should you not switch away from Scale AI?
When your volume or domain breadth is the point. Scale describes itself as supplying data to frontier model builders and lists enterprise, government and defense work on its site. None of the European tools in this directory publishes anything comparable in scale, and the managed ones (Kognic, isahit, Innovatiana) are small by comparison. A lab that needs thousands of expert contributors next week should be honest about that.
It also makes little sense to move when the data is not sensitive and the budget is tight. Public images, already licensed, labeled in bulk, do not need a European jurisdiction argument, and SentiSight.ai at 1 euro per 1,000 labels is the only European option that competes on a price you can read.
Finally, do not switch just to change the flag on an unchanged workflow. If the Scale relationship works and your legal team has accepted it, the European tools are worth a pilot for the next project, not a migration of the last one.
Which tools work for regulated or audit-heavy projects?
Kili Technology is the one that talks most directly about audit. The vendor describes a record of every version of a judgment, who made it and how reviewers agreed, and mentions the EU AI Act on its site. It also installs on your own Kubernetes with no internet or root access needed after installation, but that is an Enterprise-only add-on.
Picsellia lists ISO 27001:2022 certification on a vendor page and offers on-premise and air-gapped deployment. Keylabs claims GDPR and ISO 27001 compliance, which we label a vendor claim because its legal pages date from 2020. Isahit offers an AI Act Trust and Validate service for human oversight testing.
For medical imaging, Supervisely lists multi-planar DICOM labeling and Innovatiana lists medical data among its offers. Neither page tells you which certifications apply to your case, so ask for the documents rather than the badge.
