Argilla
Free tierThe open-source collaboration tool for AI engineers and domain experts to build high-quality datasets
Free·All audiences·Powered by Hugging Face·API available·Open source
Key strengths
Open-source and self-hostable with full data controlIntuitive API for seamless integration into existing ML pipelinesSupports RLHF, fine-tuning, and active learning workflowsCombines AI automation with human-in-the-loop feedbackStrong community support and ecosystem via Hugging Face
Completely free
Madrid, Spain
Founded 2021
Self-hostable
No ratings yet
- Supervised fine-tuning (SFT) dataset creation — Log, annotate, and export instruction-response pairs or classification datasets for LLM fine-tuning.
- RLHF preference labeling — Collect ranked or binary preference data over model outputs to train reward models at scale.
- Distilabel pipelines — Use the Distilabel library to generate synthetic training data with LLMs and validate it with human reviewers in Argilla.
- Active learning loops — Integrate Argilla with your model serving layer to surface low-confidence predictions for targeted human review.
- Hugging Face Hub integration — Push curated datasets directly to private or public Hugging Face Hub repositories with a single SDK call.
- Custom annotation workflows — Define bespoke task schemas (token classification, text ranking, multi-label, etc.) via the SDK and deploy them to annotators through the web UI.