doccano

Overview

doccano is an open-source annotation platform for turning datasets into labeled material for machine-learning work. It is best suited to teams that want control over deployment and a direct path from project setup to dataset export.

Its strongest case is a flexible, free workflow with APIs and several deployment routes; its clearest operational caution is the risk of losing a SQLite database during an upgrade.

Key features

doccano covers text classification, sequence labeling, and sequence-to-sequence tasks, with examples such as sentiment analysis, named-entity recognition, and text summarization. These task types make it a focused choice for structuring text labels rather than a general-purpose project suite.

A typical workflow moves from project creation and dataset import through adding users, defining annotation guidelines, labeling, and exporting the finished dataset. Collaborative annotation and human review workflows give teams a way to coordinate and check labeling, while custom ontologies and model-assisted labeling support more tailored projects.

The platform supports text, image, and audio modalities, alongside multi-language and mobile support, emoji support, and a dark theme. REST APIs connect it to scripts and machine-learning models. The API is useful for fitting annotation into an existing pipeline, though doccano's documented task types are specifically text-oriented.

Installation is documented through pip, Docker, and Docker Compose. The documented machine requirement is Python 3.8 or newer on Linux, Windows, or macOS; deployment options include Docker and one-click AWS and Heroku routes. Imported datasets can use Amazon S3 or Google Cloud Storage. Celery handles long-running import and export work, with SQLite3, RabbitMQ, and Redis documented as broker options.

SQLite 3 is the default database, and PostgreSQL and other database systems are also described. The upgrade warning matters: an upgrade of a SQLite installation can lose its database, so teams should treat database choice and upgrade planning as operational decisions, not afterthoughts.

Pricing

doccano is free, open-source software under the MIT license. Its listed plan costs nothing, and the license permits use, copying, modification, publication, distribution, sublicensing, and sale. That makes it an unusually permissive option for individuals and organizations that can manage their own deployment.

There are no paid tiers or plan-based seat and usage caps in the pricing model. The tradeoff is that self-hosting places setup, deployment, and database upkeep with the user rather than bundling them into a paid hosted plan.

Platforms

doccano is available for web and self-hosted use, with support across Linux, macOS, Windows, Android, and iOS, as well as API access. The documented installation routes and Python requirement make it practical for teams with a suitable machine or container environment; cloud deployment is documented for AWS and Heroku.

Who it's for

Choose doccano if you need collaborative dataset annotation, human review, and export in a free tool you can deploy and integrate yourself. It fits machine-learning practitioners and teams that value open licensing, custom label structures, and API-based workflows.

It is a weaker fit for teams that want a managed service or cannot take responsibility for database upgrades and deployment. The SQLite warning is particularly relevant for small installations that might otherwise accept the default without planning how to protect data.

Pros and cons

• Pro: The MIT license and free plan allow broad use, including modification and redistribution, without a paid tier.
• Pro: Projects, guidelines, collaborative annotation, review, and export form a complete labeling workflow rather than a standalone labeling screen.
• Pro: REST APIs, Docker deployment, and S3 and Google Cloud Storage import backends help connect it to data and model workflows.
• Con: Self-hosted deployment means the team owns installation and database decisions instead of relying on a bundled hosted service.
• Con: Upgrading a SQLite installation can lose its database, making the default choice risky without an upgrade and backup plan.
• Con: The documented task types are text-focused, even though the supported modalities also include image and audio.

Alternatives

For a broader comparison, browse AI Data Labeling Tools.

• LabelU is another free option with API, macOS, self-hosted, web, and Windows platforms; consider it if that platform mix better fits your environment.
• Label Studio has a free plan and free trial, with API, Linux, macOS, self-hosted, web, and Windows support. Its freemium model may suit readers who want to consider more than a free-only offering.
• Alibaba Cloud PAI iTAG offers manual labeling free, while OSS storage and data transfer are billed separately; choose it if that distinction fits your team's cost model.
• BRAT is a free, MIT-licensed, self-hosted option for Linux, macOS, and web; it is worth considering when that platform set meets your needs.
• Roboflow has a free tier with 10 monthly credits, enough for about 30 model trainings or 80,000 inferences, and a Core plan at $39 a month. Consider it if those model and inference allowances suit your workload.
• CVAT offers a free MIT-licensed Community plan for personal use and small teams, plus an online free plan limited to 1 member, 1 project, 3 tasks, and 1 GB. Its online limits may fit a small trial, while doccano has no plan-based caps in its free model.
• Argilla is free, open-source software deployable on Hugging Face Spaces or your own infrastructure; consider it if either deployment route is preferable.
• BasicAI offers private-cloud deployment starting at $6,600 a year, with customizable payment cycles, seats, storage, model calls, and on-premise options. It suits readers specifically seeking configurable private-cloud deployment and prepared for that starting price.

Verdict

doccano is a strong choice for machine-learning teams that want free, permissively licensed annotation with collaboration, review, and export under their own deployment control. Its APIs and infrastructure options add practical flexibility. Look elsewhere if you need a managed service, and do not rely on the default SQLite setup without accounting for the documented upgrade data-loss risk.