BRAT

BRAT is a free, self-hosted tool for annotating text with structured labels. It is best suited to research groups and teams that need a configurable scheme and shared editing. Its annotation and export tools are capable, but the team must manage deployment and secure production hosting.

Overview

BRAT is built around annotations that computers can process, rather than informal highlights or comments. Teams can define categories, annotation types, and constraints for a particular project, then work across documents using the same scheme. That focus makes it a strong fit for structured text work, but not for projects that need to label images, audio, or video.

Within text, the annotation model covers spans, binary relations, n-ary associations, attributes, normalization, and freeform notes. Unicode support includes nearly 100 scripts, giving multilingual projects broad coverage. A new installation assumes a UNIX-like environment; Windows users can run it in a virtual machine with a UNIX-like operating system.

Key features

Configurable schemes and validation

Teams can tailor annotation categories, types, and constraints to their task. BRAT validates edits against those constraints, which can help catch inconsistent labels as work progresses. This is most useful when a project can settle on a clear schema; flexibility alone does not ensure that annotators apply it consistently.

Collaboration, search, and automation

Multiple annotators can work on the same collection or document and see one another’s edits nearly as they are made. Search across documents or collections, including keyword-in-context concordancing, helps teams find relevant passages without treating each file as an isolated task. BRAT can also integrate output from automatic annotation services exposed as web services, though the project must have such a service available.

Normalization and export

Normalization can connect annotations to external resources such as Freebase, Wikipedia, and Open Biomedical Ontologies. Annotations export in brat stand-off format, while visualizations can be saved as SVG, PNG, PDF, or EPS. These options suit projects that need structured annotation data or shareable visual output; the supported export formats are specific rather than a promise of compatibility with every downstream tool.

Pricing

BRAT is open-source software under the MIT License. The brat v1.3 (Crunchy Frog) plan costs nothing and requires a self-hosted installation. There is no software fee, but the team takes on installation and hosting. That trade-off favors groups with an environment and capacity to operate the service; it is less suitable for teams seeking a hosted service instead of managing deployment.

Platforms

BRAT is web-based and supports Linux, macOS, and self-hosted deployment. A new server installation assumes a UNIX-like environment, with a virtual machine running a UNIX-like operating system as the stated route for Windows users. For production, the project recommends a full web server: its experimental standalone server is not security-hardened and is not recommended for sensitive data or systems accessible from the internet.

Who it's for

BRAT suits research groups and other teams doing text annotation that can define a shared schema, collaborate on the same documents, and manage their own hosting. Model-assisted labeling, human review workflows, custom ontologies, and API access align with projects that combine structured annotation with broader labeling processes. It is not the right choice for multimodal labeling or for teams that cannot provide a suitable UNIX-like deployment and production web server.

Pros and cons

• Pro: Configurable annotation types and constraint validation support project-specific schemes and can flag edits that do not conform.
• Pro: Near-real-time shared editing, collection-wide search, and multiple export formats support collaborative text projects from annotation through review and output.
• Pro: Unicode support across nearly 100 scripts makes the tool relevant to a wide range of multilingual text work.
• Con: Self-hosting requires an operating environment suited to a UNIX-like server installation, adding deployment work for teams without that setup.
• Con: The experimental standalone server is not hardened for sensitive or internet-accessible deployments, so production use calls for a full web server.
• Con: Its scope is text only, ruling it out for image, audio, or video annotation projects.

Alternatives

AI Data Labeling Tools is a broader category directory for comparing labeling software.

• doccano is another free option, with Android, iOS, Windows, API, and other platform support as well as Linux, macOS, web, and self-hosting.
• Label Studio is worth considering if a freemium product with a free trial and Windows support better fits your deployment needs.
• VGG Image Annotator is a free alternative with Linux, macOS, Windows, and web support and no paid tiers or usage limits stated.
• Alibaba Cloud PAI iTAG may suit teams seeking a cloud labeling option; manual labeling is free, while OSS storage and data transfer are billed separately.
• Roboflow offers a freemium plan with 10 credits per month, enough to train about 30 models or run 80,000 inferences, and is an alternative for teams whose work fits that allowance.
• Amazon SageMaker Autopilot is a paid option without a free plan, with pay-as-you-go pricing and no minimum fees or upfront commitments for on-demand use.
• Argilla is another free, open-source choice that can be deployed on Hugging Face Spaces or a team’s own infrastructure.
• CVAT has a free Community plan for personal use and small teams, plus a free online tier limited to one member, one project, three tasks, and 1 GB.

Verdict

Choose BRAT if your team needs collaborative, configurable text annotation and can operate its own production server. Its strongest case is the combination of constraint validation, shared editing, search, and structured exports at no software cost. Look elsewhere if you need multimodal labeling or a managed deployment, or if you cannot provide a full web server for production.