Argilla

Argilla is an open-source (Apache 2.0) data annotation and curation platform for AI engineers and domain experts building datasets for LLMs and NLP models. Acquired by Hugging Face in 2024 in a roughly $10 million deal, it is now deeply integrated with the Hugging Face ecosystem while remaining independently usable.

The platform combines hand labeling with active learning, bulk labeling, and AI-assisted suggestions, supporting RLHF and preference-tuning pipelines, RAG feedback collection, classification, NER, and multimodal datasets. Engineers drive it through a Python SDK or a FastAPI-based REST API, and annotators work in a purpose-built UI with filters, semantic search, and AI feedback suggestions. It deploys with a single Docker command and has been used to produce notable open datasets such as the cleaned UltraFeedback preference set. The main caveat is operational: it is a full application (UI, API, search index) rather than a library, so small teams should expect real hosting overhead.

Self-host with docker run -d --name argilla -p 6900:6900 argilla/argilla-quickstart:latest, or deploy on Hugging Face Spaces. Install the Python SDK with pip install argilla.