Doccano is a widely used open-source annotation tool for building labeled datasets for machine learning. It gives teams a clean web interface to label text for classification, sequence labeling and sequence-to-sequence tasks, and has collected about 10,800 stars on GitHub with development continuing into 2026.
Annotation is the unglamorous bottleneck of most NLP projects, and Doccano removes most of its friction. You import raw text in formats like JSONL, CSV or plain text, define your label scheme, and annotators work through examples with keyboard-friendly tooling for document classification, named entity recognition with span highlighting, and text generation tasks. Multiple users can collaborate on the same project with role-based access, progress tracking and agreement checks, and finished datasets export in standard formats ready for training. An auto-labeling feature can call external model APIs to pre-label examples so humans only review and correct. Doccano runs anywhere with Docker, entirely under your control, which makes it suitable for sensitive corpora. Doccano’s longevity is part of its appeal: it is one of the most deployed open-source labeling tools in the world, with an official Docker image, one-command deployment guides for major clouds and a large community answering usage questions. Projects can enforce annotation guidelines, randomize example order and restrict label sets per task, and a REST API lets you automate dataset management inside larger pipelines. For confidential text — legal, medical or internal documents — self-hosting keeps everything inside your own network.
Doccano is completely free and open source under the MIT license, with no per-seat or per-label fees of any kind.
Doccano is for ML teams, researchers and data operations groups who need reliable, self-hosted text annotation — from NER datasets to LLM fine-tuning and evaluation data.