- Home
- AI Data Labeling Tools
- Prodigy
Prodigy
Berlin, Germany
What it is
Prodigy is a downloadable developer tool and Python library for creating training and evaluation data for machine-learning models. Built-in recipes support named entity recognition, text classification, computer vision, audio and video annotation, transcription, and model training and evaluation. Its customizable web application provides 23 interfaces covering text, images, audio, video, relations, multiple choice, HTML, and conflict resolution. Workflows can be defined with Python recipes, while custom interfaces use HTML, CSS, and JavaScript. Prodigy supports spaCy, Hugging Face model plugins, and major LLM API providers. Data can be stored in SQLite, MySQL, or PostgreSQL and exported to JSON or JSONL. It runs on users’ machines, can operate air-gapped, and can be deployed through its Python server in the cloud or on-premise. The Personal license costs $390 once; Company costs $490 per seat once in five-seat packs, both excluding tax. Hosted trials are for companies and organizations, not individuals.
For and against
For
- 23 customizable annotation interfaces
- Python workflow and interface customization
- spaCy and Hugging Face integrations
- Air-gapped operation supported
- JSON and JSONL exports
Against
- No free plan
- Personal license costs $390 once
- Company license requires five-seat packs
- Hosted trials exclude individuals
The verdict
Prodigy combines broad annotation interfaces with programmable workflows, local operation, and integrations for machine-learning development. Its paid licenses and developer-oriented setup are clearest for organizations or technically comfortable users.
Who it suits
Prodigy suits developers and data teams building labeled datasets for machine-learning systems. Basic Python and command-line familiarity is expected, although annotators can use the web app after setup.
Plans and prices
Personal
12 months free upgrades · unlimited personal and professional use · one individual license
Read 21 Oct 2026 · prodigy.ai
Company
lifetime license · 12 months free upgrades · transferable company seats · HTTP authentication and OpenID Connect SSO
Read 21 Oct 2026 · prodigy.ai
On the record
- Active learning
- Active-learning workflows are provided for named-entity recognition, text classification, part-of-speech tagging and dependency parsing.prodigy.ai · read 21 Oct 2026
- Annotation interfaces
- The product includes 23 customizable interfaces for text, images, audio, video, relations, multiple choice, HTML and conflict resolution.prodigy.ai · read 21 Oct 2026
- Annotation tasks
- Built-in recipes cover tasks including named entity recognition, text classification, computer vision, audio and video annotation, transcription, and model training and evaluation.prodigy.ai · read 29 Sep 2026
- Annotation workflows
- Built-in workflows cover named-entity recognition, span categorization, text classification, part-of-speech tagging, dependency parsing, coreference, computer vision, audio, video, transcription and speaker diarization.prodigy.ai · read 21 Oct 2026
- Company headquarters and founding
- LinkedIn lists Explosion as headquartered in Berlin, founded in 2016 and employing 2–10 people.linkedin.com · read 21 Oct 2026
- Company legal name
- The maker is ExplosionAI GmbH, registered in Berlin, Germany.explosion.ai · read 21 Oct 2026
- Custom front end
- The interface can be extended with custom HTML, CSS and JavaScript and by combining built-in interface components.prodi.gy · read 21 Oct 2026
- Customization
- Users can define workflows with Python recipes and build custom interfaces using HTML, CSS, and JavaScript.prodigy.ai · read 29 Sep 2026
- Data formats
- Built-in loaders accept JSONL, JSON, CSV, plain text and image directories.prodigy.ai · read 21 Oct 2026
- Deployment
- The standard Python web server and application can be deployed on any cloud provider or entirely on-premise.prodigy.ai · read 29 Sep 2026
- Documentation and support
- The maker provides extensive online documentation, examples and a community support forum; company plans add priority forum and email support.prodigy.ai · read 21 Oct 2026
- Extensibility
- Custom Python recipes can define data streams, preprocessing, model callbacks, annotation interfaces and training logic.prodigy.ai · read 21 Oct 2026
- Funding and ownership
- Explosion says SignalFire invested $6 million for 5% in 2021, invested another $3 million in 2022 and retained 7.7% after the company returned to self-sufficient operation in 2024.explosion.ai · read 21 Oct 2026
- Hugging Face integration
- The Prodigy-HF plugin can train transformer models, use Hugging Face checkpoints for correction workflows and upload datasets to the Hub.prodigy.ai · read 21 Oct 2026
- Installation
- Licensed users install through pip using a private download index and license key, or install a downloaded wheel.prodigy.ai · read 21 Oct 2026
- Integrations
- Prodigy has first-class support for spaCy, plugins for Hugging Face models, and out-of-the-box integrations with major LLM API providers.prodigy.ai · read 29 Sep 2026
- Intended users
- Prodigy is designed as a developer tool and assumes basic familiarity with Python and the command line, while its annotation web app can be used without programming experience after setup.prodigy.ai · read 29 Sep 2026
- Licence
- Prodigy is proprietary commercial software; the licence is revocable, non-exclusive and non-transferable, with some bundled components potentially under open-source licences.prodigy.ai · read 21 Oct 2026
