Argilla

Argilla is the tool ML teams reach for when they realize 'better data beats a better prompt'. It is an open-source platform where domain experts, annotators, and engineers collaborate to label, rate, and curate the datasets that train and evaluate language models. You can collect human feedback (preference rankings, ratings, free-text critiques) on model outputs, build supervised fine-tuning datasets, run RLHF/DPO data collection workflows, and continuously monitor production model quality by sampling responses for review. Acquired by Hugging Face in 2024, Argilla integrates natively with the Hugging Face Hub, datasets library, and AutoTrain — making it the default labeling layer for the open-source LLM ecosystem. The Python SDK lets engineers programmatically push records, set up annotation guidelines, and sync results, while the web UI gives non-technical reviewers a clean, keyboard-driven labeling experience. Argilla is free and open source (Apache 2.0); you can self-host it locally with Docker, deploy on the Hugging Face Spaces in one click, or run it on your own Kubernetes cluster. It is widely used by teams building domain-specific or multilingual LLMs where the bottleneck is data quality, not compute.