Autolabel is an open-source Python library from Refuel AI with 2,300+ stars on GitHub that uses large language models to label, clean and enrich text datasets — the training data work that usually eats weeks of manual annotation.
Clean, diverse labeled data is the bottleneck of most machine learning projects, and modern LLMs such as GPT-4 can label data with high accuracy at a fraction of the cost and time of manual labeling. Autolabel, actively maintained and last updated in 2025, turns that idea into a practical library: you define a labeling configuration, point it at your raw text dataset, and it orchestrates LLM calls to classify, tag, extract or enrich each record. It works with the LLM of your choice — OpenAI models, Anthropic Claude, Hugging Face transformers and LangChain-backed providers — and keeps labeling runs structured, resumable and auditable. Refuel also publishes benchmark tooling around the library so teams can measure model labeling quality before committing to a full run, making it a solid building block for data-centric AI workflows.
Autolabel itself is free and open source under the MIT license; your only library-level cost is the LLM API usage you choose to spend. Refuel AI also offers a hosted platform with its own plans for teams that want a managed data labeling service.
ML engineers and data teams who need large, clean labeled text datasets for training or evaluating models and want LLM automation instead of slow, expensive manual annotation.