HumanSignal

HumanSignal builds Label Studio, the open-source tool teams use to label their own data, and sells the hosted and enterprise version around it.

The open-source edition is the way most people arrive. You install it, describe the labeling interface in a short config, and your annotators start working. It is Apache 2.0 licensed, so there is no seat cost and nothing stops you running it on your own hardware with data that never leaves your network. That is a large part of why it turns up in research groups and in companies whose legal team will not approve an outside cloud.

One tool covers formats that usually take several:

• Images for classification, detection, tracking, and segmentation
• Audio for transcription, speaker diarization, and emotion tagging
• Text for entity recognition, sentiment, and question answering
• PDFs and scanned documents, including OCR
• Video
• Time series and sensor streams
• Multimodal projects that put several of these in one view

You can connect a model to a project and have its predictions load as pre-annotations, so annotators correct output rather than start from a blank frame. An API, a Python SDK, and webhooks handle the plumbing around that, from pushing tasks in to pulling finished labels back out.

Label Studio Enterprise is what teams move to once the project outgrows a handful of annotators. It adds:

• More than 30 agreement metrics, including IoU for boxes and polygons, edit distance for transcription, and F1 for classification
• Dashboards showing time per task and accuracy per annotator
• Review workflows and comparison against ground truth
• Role-based access, SSO, and SCIM, which matter once contractors touch the data
• Hosting on HumanSignal's cloud or a deployment inside your own infrastructure

Newer work leans toward evaluation: comparing model outputs side by side against a rubric your team writes, so you benchmark on your own criteria instead of a public leaderboard.

HumanSignal added a services arm in late 2025 after acquiring Erud AI. That side recruits and manages expert AI trainers for labeling, evaluation, and red teaming, and it runs collection projects, including on-site capture for audio, vision, and robotics data. So the same vendor can supply the tooling, the people, or both.

The company was called Heartex until it renamed to HumanSignal in 2023, worth knowing if you hit the old name in older docs and comparison posts.

Compliance covers SOC 2 and HIPAA, and self-hosting is there for data that cannot sit on someone else's cloud at all.

Free and open source for the community edition. The enterprise edition is quoted per organization by seats, deployment model, and support.

Best for teams that want to run labeling themselves, especially ones juggling several data types in one tool or required to keep the whole thing inside their own network. Not the right fit for a team that would rather hand over the data and get finished labels back without running any of the process, where a managed-workforce vendor is the closer match.