doccano

doccano is a free, open-source tool for labeling data for human review and machine-learning work. It supports text classification, sequence labeling, and sequence-to-sequence tasks, with examples including sentiment analysis, named-entity recognition, and summarization. A project workflow lets users create projects, import datasets, add collaborators, set guidelines, annotate data, and export labeled results. Collaboration features include multi-language and mobile support, emoji support, and a dark theme. The software supports text, image, and audio modalities, model-assisted labeling, human review workflows, and custom ontologies. REST APIs connect scripts and machine-learning models. doccano can be self-hosted; installation is documented through pip, Docker, or Docker Compose on Linux, Windows, and macOS machines running Python 3.8 or newer. Deployment options include AWS, Heroku, or Docker elsewhere. Imported datasets can use Amazon S3 or Google Cloud Storage. SQLite 3 is the default database, with PostgreSQL and other systems also documented. Celery handles long-running import and export tasks. The installation guide warns that upgrading a SQLite3 installation can result in database loss.

doccano combines annotation workflows, collaboration, and API access in a free, self-hosted tool. Users upgrading a SQLite3 installation should account for the documented risk of database loss.

doccano suits teams and practitioners who need to label data for machine-learning or human review workflows. It is also relevant to users comfortable with self-hosting and its documented installation options.

Overview

doccano is a self-hosted labeling workspace for organizing data annotation with a team. It suits practitioners who want to manage projects and move labeled datasets into scripts or machine-learning workflows. Its appeal is a no-cost, configurable workflow; deployment and database upkeep remain the user's responsibility.

Key features

Projects can take a dataset from import through user assignment, annotation guidelines, labeling, and export. That end-to-end structure is useful when a team needs shared conventions and a handoff to downstream work, rather than a standalone labeling step. Collaborative annotation and human review workflows support team coordination, while custom ontologies and model-assisted labeling give projects more structure and assistance.

The documented task types are text classification, sequence labeling, and sequence-to-sequence work, with examples including sentiment analysis, named-entity recognition, and text summarization. Although supported modalities include text, image, and audio, the stated task types center on text. Teams choosing it for image or audio work should account for that distinction.

REST APIs connect doccano to scripts and machine-learning models, which makes it more adaptable to an existing pipeline than a purely manual workflow. Multi-language support, mobile support, emoji support, and a dark theme round out the interface and collaboration features.

Installation is documented through pip, Docker, and Docker Compose. The project also documents one-click deployment options for AWS and Heroku, and Docker deployment elsewhere. Linux, Windows, or macOS machines running Python 3.8 or newer meet the documented runtime requirement. Imported datasets can use Amazon S3 or Google Cloud Storage. SQLite 3 is the default database, with PostgreSQL and other database systems also described; Celery handles long-running imports and exports, with SQLite3, RabbitMQ, and Redis documented as broker options. These choices offer deployment flexibility, but require operators to make and maintain infrastructure decisions.

One operational warning deserves attention: the installation documentation cautions that upgrading a SQLite3 installation can lose its database. The repository has no detected SECURITY.md policy and no published security advisories. Neither point establishes a security failure, but organizations should assess their own security and upgrade requirements before adopting a self-managed installation. Users who get stuck are directed to the FAQ and can send help requests or feedback to the author.

Pricing

MIT-licensed software costs nothing, with permission to use, copy, modify, publish, distribute, sublicense, and sell. There are no paid tiers or seat and quota limits described. That makes doccano suitable for individuals and teams seeking software they can adapt without a subscription; in exchange, they take on deployment and maintenance rather than buying a managed service.

Platforms

doccano supports Android, iOS, Linux, macOS, Windows, web, API, and self-hosted use. The platform list is broad, while the documented installation paths and runtime requirement make clear that running it involves setting up an environment rather than simply selecting a hosted plan.

Who it's for

Choose doccano when text-labeling projects need team collaboration, review, configurable annotation structure, dataset export, and API integration, especially when self-hosting is a requirement. It is a less natural fit for buyers seeking a managed service, or for teams that need modality-specific workflows beyond its documented text task types. Operators using SQLite3 should plan upgrades carefully.

Pros and cons

• Pro: The full project-to-export workflow, collaborative annotation, and human review suit teams coordinating labeled datasets.
• Pro: REST APIs, multiple installation routes, and database and broker options allow integration and deployment choices.
• Pro: MIT-licensed software is free to use and modify, with no stated seat or usage caps.
• Con: Self-hosting puts installation, infrastructure, and ongoing operation on the user.
• Con: SQLite3 upgrades carry a documented risk of database loss, so a default installation requires care during upgrades.
• Con: The stated task types are text-focused, which may not satisfy teams looking for clearly defined image or audio labeling tasks.

Alternatives

For a wider set of AI data-labeling options, browse AI Data Labeling Tools.

Roboflow may suit teams prioritizing model training and inference: its free tier includes 10 credits a month, described as enough to train about 30 models or run 80,000 inferences.

CVAT is worth considering for a limited personal or small-team setup under its MIT-licensed Community plan, or a hosted free plan capped at one member, one project, three tasks, and 1 GB.

Alibaba Cloud PAI iTAG is an option for teams doing manual labeling for free, provided they account separately for OSS storage and data-transfer charges.

Amazon SageMaker Autopilot fits readers seeking pay-as-you-go use without minimum fees or upfront commitments, with a free trial available.

BRAT is another free, MIT-licensed, self-hosted option.

Intellisane Annotation Platform may suit a small project that can stay within its free allowance of three members, two projects, 1,000 annotations, and 200 assets.

Label Studio is another freemium option with a free plan and trial.

Supervisely may fit a small team that can work within the Community plan's 5 GB storage, 10,000 files, and two members, with unlimited projects and annotations.

Verdict

doccano is a strong choice for machine-learning practitioners and teams who want a free, self-hosted annotation workflow with collaboration, review, export, and API access. Its flexibility and open license are the main reasons to choose it; teams that do not want to operate their own deployment, or cannot accept the SQLite3 upgrade risk, should look elsewhere.