Skip to content
EN
English 简体中文 soon 日本語 soon

Label Studio

Open-source annotation platform

Visit official site

What Label Studio is

Label Studio is an open-source platform for creating the datasets everything else depends on. It supports annotation across computer vision, document AI, NLP and audio, and newer uses include agent trajectory review, model evaluation and preference data collection.

What you can do with it

  • Configure labelling interfaces per task
  • Annotate images, text, audio and documents
  • Review agent trajectories
  • Collect preference and evaluation data
  • Export in formats training pipelines accept

Who it is for

  • Machine learning and data teams
  • Researchers building evaluation sets
  • Organisations running annotation programmes

What to watch out for

  • Annotation quality depends on instructions and agreement between annotators, not on the tool
  • Human annotation has labour implications: pay and brief annotators properly, and do not treat disagreeing labels as noise
  • Data you load may contain personal information, which needs access control and retention rules
  • Self-hosting means upgrades, backups and security patching are yours

Pros & cons

✓ What we like

  • Genuinely open source
  • Very broad data type coverage
  • Usable for evaluation as well as training

! What to watch out for

  • Instruction quality decides data quality
  • Annotator welfare and clarity matter
  • Self-hosting carries maintenance

FAQ

What data types are supported?

Computer vision, document AI, NLP, audio transcription, agent trajectories and evaluation tasks.

Is there a hosted version?

Open-source and hosted options are both offered.

What determines dataset quality?

Clear instructions, annotator agreement and review, not the tool itself.

Last reviewed: 2026-09-19

More LLM API platform tools

View all →

How we review